Energizing the Machine
What Mitochondria Can Teach Us About Powering the Next Generation of Robots
Large batteries are the wrong answer. Nature solved this problem 2 billion years ago.
By Steven Muskal, Ph.D. | The Rise of the Machine II| stevenmuskal.com
The centralized battery paradigm: hundreds of kilograms of electrochemical cells that define and constrain modern electric vehicles.
The Crushing Weight of Our Best Ideas
Here’s a number that should unsettle you: 480 kilograms.
That’s the mass of the battery pack in a Tesla Model S Long Range. Nearly half a metric ton of electrochemical cells bolted to the floor of what we’ve been told is the future of transportation. To put it viscerally: you’re hauling around the weight of a baby elephant every time you drive to the grocery store.
And it’s not just Tesla. The Rivian R1T pickup carries a 800 kg battery. The GMC Hummer EV? A staggering 1,326 kg of battery alone, more than many entire compact cars weigh. These vehicles have become architectural monuments to a single engineering decision: centralized energy storage.
The consequences ripple outward with punishing physics. Heavier batteries demand reinforced suspension systems. Reinforced suspensions require beefier brakes. Beefier brakes add mass. More mass requires more energy to accelerate and decelerate, which demands… more battery capacity. We’ve engineered ourselves into a feedback loop where our solution to the energy problem is the energy problem. Recent research has confirmed that EV-induced road damage scales dramatically with axle load, following the well-established fourth-power law[1].
But here’s what keeps me awake at night, and what I’ve been building toward across this Rise of the Machine series: we’re about to make this same mistake with humanoid robots.
Boston Dynamics’ Atlas. Tesla’s Optimus. Figure AI’s 01. The next generation of AI-embodied systems. Every one of them faces the same existential constraint: mass is the enemy. A robot that must carry a 50 kg battery pack to operate for two hours is a robot that has already lost. Every gram allocated to energy storage is a gram stolen from payload capacity, from actuator power, from the sensors and compute that make intelligence possible.
What if the entire paradigm is wrong? What if the answer isn’t “better batteries” but no batteries at all, at least not in the way we’ve conceived them?
Nature solved this problem two billion years ago. And she didn’t use lithium.
Section 1: The Mass Problem Scales Catastrophically
Let’s sit with the physics for a moment, because the numbers are damning.
In modern electric vehicles, battery mass typically represents 25–30% of total vehicle weight. The Tesla Model 3’s battery accounts for approximately 28% of its curb weight. For the Hummer EV, batteries approach 35%. These aren’t engineering compromises; they’re structural commitments that cascade through every other design decision.
Consider what this means mechanically. A 2024 study on hidden externalities of electric vehicles documented that EVs impose significantly greater wear on road surfaces and tire systems precisely because of this concentrated mass[2]. The fourth-power law of road damage, which states that wear scales with the fourth power of axle load, means even modest weight increases impose dramatically disproportionate consequences. Suspension components in EVs show accelerated wear patterns. Brake systems require oversizing. Tire compounds must be reformulated to handle the additional loading.
Now scale this to a humanoid robot.
A robot operating in a warehouse, a hospital corridor, or alongside humans in a factory faces constraints far more severe than any vehicle. Every additional kilogram is:
- More torque demand on every joint actuator
- Greater inertial load during acceleration and deceleration
- Increased ground reaction forces transmitted through the structure
- Larger actuators required, which themselves add mass
- More energy consumed per movement cycle
This is the compounding mass penalty, and it’s why battery weight isn’t just inconvenient for robots; it’s architecturally fatal.
Current humanoid prototypes allocate 20–40% of their mass budget to batteries. Some early systems were effectively battery packs with legs. The operational consequences are predictable: limited runtime (often under two hours), restricted payload capacity, and compromised dynamics.
Core Insight
In mobile systems, energy storage and structural mass are in zero-sum competition. Every kilogram devoted to storing energy is a kilogram that cannot contribute to capability. Biology never makes this trade-off. And that should tell us something profound.
The Caloric Reality Check: Human vs. Machine
Before we can solve the power problem, we need to quantify it. Let’s put human and robot energy demands side by side.
An active adult human consumes approximately 2,000–2,500 kilocalories per day. Converting to physics: 2,000 kcal = 8.4 megajoules, spread across 86,400 seconds, yields roughly 97 watts of continuous average power. During 16 waking hours of movement, cognition, and homeostasis, the effective draw rises to approximately 145 watts average. Peak output during sprinting can reach 1,500–2,000W, but only for seconds.
From this modest ~100W continuous budget, a 70 kg human walks, thinks, maintains body temperature, repairs tissues, and runs a 20-watt brain. The secret: biological muscles are only ~25% mechanically efficient (75% becomes heat), but they excel at passive holding via tendons (zero energy to maintain position) and elastic energy recovery.
The striking finding: despite being 3–4x more efficient electrically, the robot requires 3–4x more absolute power than a human because electric actuators lack muscles’ passive holding ability, and because the robot has no equivalent of tendons that store and release energy for free. The daily energy budgets are remarkably similar (~8.4 MJ for both), but the human achieves continuous operation through constant metabolic intake while the robot must periodically stop and recharge.
This equivalence is not a coincidence. It tells us that a 57–70 kg mobile system performing human-equivalent work in Earth’s gravity well requires approximately 8–10 megajoules per operational day, regardless of whether the prime mover is muscle or motor. The question isn’t “how much energy?” but “how do you deliver it?”
Fluorescence microscopy reveals the intricate internal architecture of living cells. The colored organelle networks visible here include mitochondria, the distributed energy systems refined over two billion years of evolution. Mitochondria: not merely powerhouses, but distributed, adaptive, intelligent energy management systems refined over two billion years of evolution. See Mitochondria and future of medicine.
Section 2: What Biology Figured Out
If I were to ask you what mitochondria do, you’d probably answer: “They’re the powerhouse of the cell. They make ATP.”
You’d be right. And you’d be missing almost everything interesting.
Mitochondria are not simply ATP factories. They are distributed, adaptive, intelligent energy management systems that also serve as signaling hubs, quality control monitors, and, remarkably, executioners. Understanding what mitochondria actually do is the key to understanding what our machines are missing.
Two Billion Years Ago: Energy Came First
Here is a fact that should reframe your entire conception of how complex life arose: mitochondria were not built by cells. They were free-living bacteria that invaded them.
Approximately two billion years ago, an alpha-proteobacterium, a free-living energy-producing microbe, was engulfed by a primitive archaeal host cell. Rather than being digested, it survived. A symbiosis formed. The bacterium provided a massive surplus of ATP; the host provided raw materials and protection. Over eons, the bacterium lost its independence, shedding most of its genome into the host nucleus, but it retained its own DNA, its own replication machinery, and its own double membrane, vestiges of that ancient merger.
This is the endosymbiotic theory, now supported by overwhelming molecular evidence: mitochondria retain their own circular DNA (like bacteria), replicate by binary fission (like bacteria), and are bounded by a double membrane (the inner membrane being the original bacterial membrane). They are, in a very real sense, domesticated bacteria living inside every cell of your body.
Nick Lane’s landmark work (The Vital Question, 2015; “How energy flow shapes cell evolution,” Current Biology, 2020) establishes the critical causality: the energy source preceded and enabled cellular complexity, not the reverse. Without the 100,000-fold energy surplus that internalized mitochondria provided per gene, eukaryotic cells could never have evolved nuclei, organelles, multicellularity, or any of the machinery we associate with complex life. Every complex organism on Earth, every animal, plant, and fungus, carries mitochondria. There are no exceptions. This is not coincidence; it is causation.
Two further observations deepen the insight:
Maternal inheritance: In humans and most animals, mitochondrial DNA is inherited almost exclusively from the mother. The mitochondria in your cells are descendants of those in your mother’s cells, which were inherited from her mother, and so on through a long maternal lineage. This pattern of inheritance allows mitochondrial DNA to be used to trace maternal ancestry across many generations. The mitochondria present in modern organisms ultimately descend from an ancient symbiotic relationship between an ancestral cell and a bacterium that occurred roughly 1.5 to 2 billion years ago, although mitochondrial genomes and functions have evolved substantially since that time.
The lesson for machines: The evolution of complex cells was closely linked to the acquisition of mitochondria, which greatly increased the amount of energy available to support cellular processes. Many researchers view this increase in energy capacity as an important factor enabling the evolution of larger genomes, greater cellular complexity, and multicellular organisms. In contrast, robotics and artificial intelligence development often emphasize advances in computation, sensing, and mechanical capability while relying on incremental improvements in energy storage and power systems. This comparison suggests that breakthroughs in energy generation, storage, and distribution may be as important to future robotic capabilities as advances in hardware and software.
Beyond ATP: The Hidden Lives of Mitochondria
Calcium signaling and buffering: Mitochondria actively absorb and release calcium ions, modulating the calcium waves that propagate through cells during signaling events. They’re not passive bystanders to cellular communication; they’re active participants, shaping the temporal dynamics of signals that control everything from muscle contraction to neurotransmitter release.
Programmed cell death (apoptosis): Here’s a fact that should reframe your entire conception of these organelles: mitochondria decide when cells die. Through the release of cytochrome c and other pro-apoptotic factors, mitochondria trigger the cascade that dismantles a cell from within. They are quality control systems with executive authority.
Reactive oxygen species (ROS) as signals: We’ve long thought of ROS as purely damaging, byproducts of metabolism that cause cellular aging. Recent research reveals a more nuanced picture: mitochondria use controlled ROS production as a signaling mechanism, communicating metabolic state to the rest of the cell.
Fission and fusion dynamics: This is where things get truly remarkable. Mitochondria are not static structures. They continuously divide (fission) and merge (fusion) in response to cellular energy demands. When energy demand is high and sustained, mitochondria fuse into elongated networks that maximize efficiency. When demand is localized or the organelles are damaged, they fragment for targeted repair or removal. A 2024 review in Nature Reviews Molecular Cell Biology characterized these molecular mechanisms of mitochondrial dynamics in comprehensive detail[3].
Physical migration: Mitochondria move. They travel along cytoskeletal tracks to position themselves precisely where energy is needed. In neurons, mitochondria migrate to synapses during periods of high activity. In migrating cells, they relocate to the leading edge where ATP demand is greatest.
The Molecular Turbine
At the heart of mitochondrial ATP production sits what I consider the most elegant nano-machine evolution has ever produced: ATP synthase.
This protein complex is literally a rotary motor. The F0 component, embedded in the inner mitochondrial membrane, channels protons down their concentration gradient. This proton flow drives the rotation of a central shaft (the gamma subunit) at speeds approaching 100–150 revolutions per second. The rotating shaft induces conformational changes in the F1 component that catalyze the synthesis of ATP from ADP and inorganic phosphate.
Think about this: inside your cells, right now, are trillions of molecular turbines spinning hundreds of times per second, powered by proton currents, generating the universal energy currency that allows you to read these words, move your eyes, and form thoughts about what you’re reading.
A 2023 study in Nature Communications elucidated the actual mechanism by which ATP hydrolysis couples to rotation in bacterial ATP synthase, revealing a sophisticated molecular ratchet mechanism[4]. This isn’t metaphorical engineering; it’s literal engineering at nanometer scales, refined over billions of years.
Video Resources: ATP Synthase in Action
These videos provide outstanding visual reference for the molecular machinery discussed above.
- Molecular Machines — ATP Synthase: The Power Plant of the Cell (youtube.com/watch?v=XI8m6o0gXDY)
- Describes ATP synthase as a “high-tech micromolecular power generator” with a barrel-shaped rotator of 10–15 protein subunits.
- ATP Synthesis | HHMI BioInteractive Video (youtube.com/watch?v=CSrtewCJbpg)
- HHMI BioInteractive production showing mitochondria as “electrochemical batteries that convert energy from food and oxygen into ATP.”
- ATP in Use | HHMI BioInteractive Video (youtube.com/watch?v=vUC5njkOq7o)
- Shows ATP in action powering cellular machinery: calcium ion pumps, muscle contraction, nerve transmission, gene regulation.
The Metabolic Cascade
Here’s another number that should reshape your thinking: 4 grams.
That’s approximately how much glucose is in your bloodstream at any given moment. Four grams, less than a sugar packet, circulating through roughly five liters of blood. From this seemingly minimal reserve, your body powers every thought, every heartbeat, every breath.
How? Because the body never stores energy proportional to its demand. Instead, it maintains a sophisticated metabolic cascade:
- Ingestion: Complex carbohydrates, fats, and proteins enter the gut
- Digestion: Macronutrients are broken down into absorbable components
- Distribution: Glucose enters the bloodstream, maintained at tight homeostatic levels
- Cellular uptake: Glucose enters cells and is processed through glycolysis
- Oxidative phosphorylation: Pyruvate enters mitochondria, feeding the citric acid cycle and electron transport chain
- ATP synthesis: Those spinning molecular turbines convert the proton gradient into ATP
- Immediate use: ATP is consumed within seconds to minutes of its production
Your body produces approximately 40 kilograms of ATP per day. But you never store 40 kilograms of ATP. The total ATP content of the human body at any moment is roughly 250 grams. The system works by continuous generation and immediate consumption, not bulk storage.
This is the design principle we’ve ignored: flow, not reservoir.
Supercapacitors and micro-electronic energy storage: the technological building blocks of a distributed power architecture.
Section 3: The Mitocapacitor Vision
Let me introduce a concept I’ve been developing that I call mitocapacitor architecture: a distributed energy storage and delivery system for robots that mirrors the biological strategy of mitochondria.
The foundational technology already exists. It’s called the supercapacitor.
The Supercapacitor Difference
Supercapacitors (also called ultracapacitors) store energy through electrostatic charge separation rather than electrochemical reactions. This distinction has profound implications:
Look at those numbers carefully. Yes, supercapacitors store far less energy per kilogram. But they deliver energy 10–40 times faster. They can charge in milliseconds. They last through hundreds of thousands of cycles without degradation.
We’ve been asking the wrong question. The engineering community has been obsessed with: “How do we store more energy in one place?” The biological question is different: “How do we distribute the right amount of energy, instantly, exactly where it’s needed?”
Structural Integration
Carbon fiber weave: the distinctive crosshatch pattern of woven carbon fiber composites that can simultaneously serve as load-bearing structure and energy storage medium. Research at Chalmers, Cambridge, and MIT is making this dual function a reality.
Here’s where it gets interesting. Recent research from Cambridge, MIT, and Chalmers University has demonstrated structural supercapacitors, carbon fiber composite materials that simultaneously serve as load-bearing structures and energy storage devices.
A 2024 review on carbon fiber electrodes for composite structural supercapacitors documented preparation and modification methods for creating these dual-function materials[5]. Chalmers University of Technology has demonstrated that carbon fiber structural batteries could reduce EV weight by 25 percent while maintaining mechanical integrity[6].
Imagine a robot whose skeletal frame is its distributed energy storage system. Not a battery pack bolted to a chassis, but energy storage woven into the very structure that provides mechanical support. Every strut, every beam, every joint housing becomes a local energy reserve.
This concept has already been validated in robotics. A landmark 2020 paper in Science Robotics demonstrated “biomorphic structural batteries for robotics,” showing that structural zinc batteries integrated into the robot body could provide up to 72 times more energy than standalone cells by eliminating dedicated battery enclosures[7].
This is the mitocapacitor vision: thousands of micro and supercapacitors distributed throughout the structural frame of a robot, embedded in the architecture itself, as mitochondria are embedded in muscle fibers.
AI-generated illustration of ultrafast energy storage and power delivery via electrostatic microcapacitors directly integrated on-chip for next-generation microelectronics.
There’s a further advantage biology discovered: distributed energy conversion distributes heat. A centralized battery pack concentrates thermal load, demanding liquid cooling systems that add mass and failure modes. Thermal runaway, where one overheating cell ignites its neighbors, remains lithium-ion’s existential risk. A mitocapacitor architecture eliminates this by design. Thousands of small energy nodes, each generating minimal heat, turn the robot’s structural frame into a passive radiator. The same principle keeps your body at 37 degrees: trillions of mitochondria each producing tiny amounts of heat, collectively maintaining homeostasis without any centralized cooling system.
a 3D reconstruction of the brain’s dense vascular network showing how distributed blood flow handles thermal regulation
Consider the brain: 20 watts of continuous power dissipated within 1.4 kilograms of tissue that is exquisitely sensitive to temperature fluctuations. Neurons malfunction with even small thermal deviations. Yet the brain never overheats because biology distributes both energy conversion and cooling. Dense vascular networks carry heat away from every cubic millimeter of neural tissue simultaneously. There is no centralized radiator, no heat pipe, and no fan. The architecture is the cooling system. A mitocapacitor robot inherits this principle: by distributing energy conversion across its entire frame, it eliminates thermal hot spots and the heavy cooling infrastructure they demand.
Micro-Supercapacitors: The Building Blocks
A 2020 review in npj Flexible Electronics on “Recent developments of advanced micro-supercapacitors” documented extraordinary progress in fabricating micro-scale energy storage devices[8]. These devices can be printed onto flexible substrates, embedded in composites, and integrated into remarkably compact systems.
Get Steven Muskal’s stories in your inbox
Join Medium for free to get updates from this writer.
Meanwhile, 2D MXene-based supercapacitor electrodes (two-dimensional transition metal carbides) are pushing energy densities higher while maintaining the instantaneous charge/discharge characteristics that make distributed architectures viable[9].
The ATP Pool Analog
The microcapacitor array isn’t a battery in the traditional sense. It’s an always-ready local energy reserve, continuously topped off by a slower baseline source. This mirrors the ATP pool in biological cells:
- Mitochondria maintain a local ATP concentration
- ATP is consumed within milliseconds by nearby enzymes and motors
- Mitochondria continuously regenerate ATP to maintain the pool
- The pool never depletes because generation matches consumption
The key insight is that peak demand and average demand are radically different numbers. A robot joint might require 500W for 100 milliseconds during a rapid movement, but only 5W averaged over a minute. Batteries are sized for peak demand. Mitocapacitor arrays are sized for average demand, with distributed capacitors handling the instantaneous peaks.
Section 4: The Bioreactor Bridge
Every architecture needs a complete energy story. For biology, that story begins with food and ends with ATP. For the mitocapacitor architecture, we need to identify what plays the role of the digestive system, the slow, continuous energy source that keeps the distributed array charged.
Schematic diagram represents a dual chamber Microbial Fuel Cell (MFC) showing bacterial biofilm formed on the anode and electricity generation from wastewater through microbial oxidation of organic matter.
Systematic presentation of microbial fuel cell (MFC) A bioelectrochemical approach based on metabolic activities inoculated bacteria, and the MFC acts as biocatalysts for the natural conversion to energy of organic substrates
Schematic drawing of an Enzymatic biofuel cells (EBFCs), consisting of a GOx-based bioanode (PDB: 3QVP — crystal structure of glucose oxidase) and a BOx-based biocathode (PDB: 2XLL — crystal structure of bilirubin oxidase from Myrothecium verrucaria).
The Charging Options
Inductive charging plates: The simplest approach treats electricity from the grid as “food,” delivered through wireless induction pads embedded in floors at workstations, assembly lines, or anywhere a robot stands for sustained periods. Figure AI’s Figure 03 already demonstrates this: charging coils in the robot’s feet allow it to step onto a wireless pad and charge at 2 kW while continuing to work. No plug, no docking station, no interruption. Anywhere a humanoid stands for more than a few minutes becomes a charging surface. This is the metabolic equivalent of grazing: continuous low-level energy intake during normal activity, not a discrete “refueling” event.
Microbial fuel cells (MFCs): These bioelectrochemical systems use bacterial metabolism to generate electricity from organic matter. A 2025 comprehensive review on “Microbial fuel cells powering robots and beyond” documents the complete historical arc from early experiments to fully autonomous robots powered by waste[10]. The theoretical framework of Gastrobotics (Wilkinson, 2000) envisioned robots that locate, harvest, and digest food to power themselves[11].
The EcoBot lineage: Perhaps the most compelling proof-of-concept comes from the Bristol Robotics Laboratory, where EcoBot-II demonstrated an artificial agent with a “natural metabolism,” using MFCs fed with flies and waste to power fully autonomous behavior[12]. EcoBot-III extended this to complete energy autonomy from wastewater. These weren’t toy demonstrations; they were engineering proofs that biological energy pathways can sustain robotic systems.
Enzymatic biofuel cells: These systems use isolated enzymes rather than whole microbes to catalyze electrochemical reactions. A 2024 review on glucose-based biofuel cells documented how enzymatic systems operating on glucose can achieve higher power densities than MFCs[13]. A 2019 study in Scientific Reports demonstrated practical glucose-reactive enzymatic biofuel cell performance for biomedical applications[14].
Hybrid bio-electrochemical systems: The 2021 review “Towards Bio-Hybrid Energy Harvesting in the Real-World” pushes the boundaries further, examining how bio-electrochemical and bio-mechanical processes can be combined for practical energy generation[15]. Stanford’s microbial battery technology demonstrates how an anode coated with exoelectrogenic microbes can ingest organic waste to produce sustained electricity[16].
The Biofuel Provocation
What if the charging station of the future isn’t a wall outlet but a bioreactor? A robot that can convert waste biomass, wastewater, or even ambient organic matter into electricity has fundamentally different operational constraints than one tethered to the grid. The EcoBot lineage proved this is physically possible. The question is whether we can scale the power densities to match the demands of dynamic locomotion. Current MFC power densities remain in the milliwatts-to-watts range per unit, but enzymatic cells are pushing higher, and the combinatorial approach (MFC + supercapacitor buffer) could bridge the gap between biological energy rates and mechanical power demands.
Running the Numbers: Could a Robot Have a Gut?
The provocation demands quantification. Let’s work through the engineering arithmetic to determine whether a bio-digestive architecture could realistically power a humanoid robot.
The power target: A lightweight (57 kg) humanoid optimized for efficiency needs approximately 200–300W continuous for mixed factory tasks. With distributed supercapacitors handling peak demands, the bioreactor only needs to provide the average trickle charge, not the instantaneous peaks. Target: ~130–200W continuous from the bioreactor.
MFC power densities (current state of art):
- Early MFCs (2000s): 0.001 W/m2 electrode area
- Modern MFCs (2024): 1–7 W/m2 (best: MXene-enhanced anodes reaching ~7 W/m2)
- At 3 W/m2 (practical average): need 44–67 m2 of electrode area for 130–200W
- Verdict: MFCs alone are volumetrically impractical for a humanoid form factor
Enzymatic biofuel cells (glucose-fed):
- Demonstrated power density: 1–10 mW/cm2 (= 10–100 W/m2)
- At 5 mW/cm2 (achievable mid-range): need 2.6–4.0 m2 for 130–200W
- A coiled tube, 3 cm diameter, 14 meters long = 1.32 m2 bare surface
- With engineered micro-villi (30x surface enhancement, mimicking intestinal architecture): effective area = 39.6 m2
- At 5 mW/cm2 on villi-enhanced surface: theoretical output = ~198W
The gut-tube architecture:
Fuel requirements: At 200W continuous output with ~40% glucose-to-electricity conversion efficiency, the system would consume approximately 430 grams of glucose per 16-hour operating day, dissolved in roughly 2–4 liters of solution. That’s less than a pound of sugar per shift, a remarkably achievable logistics burden compared to the infrastructure required for battery charging.
The hybrid architecture that emerges:
- Distributed supercapacitor array (structural, 500–800 Wh total): handles instantaneous peaks
- Enzymatic gut-tube bioreactor (torso-mounted, 2–4L): provides 100–200W continuous trickle charge
- Robot “eats” glucose solution periodically (like refueling a car, but with sugar water)
- The bioreactor never needs to match peak demand because the capacitors buffer everything
- During “sleep” (low-activity charging periods), the bioreactor tops off the full capacitor array
Critical caveat: These are theoretical upper bounds. Real-world enzyme degradation, mass transport limitations, biofouling, and thermal management will reduce practical output by 30–60%. A realistic near-term system might achieve 50–80W, insufficient alone but potentially viable as a range extender that doubles operational time between conventional charges. The path to full energy autonomy requires continued advances in enzyme stability (currently 30–90 day lifetimes before replacement) and micro-villi fabrication at scale.
The conclusion is provocative but grounded: a robotic digestive system is not thermodynamically impossible. The surface areas, power densities, and volumetric constraints are all within an order of magnitude of feasibility. The engineering challenges are fabrication, durability, and integration, not physics.
The Complete Biological Mapping
The Central Insight
The body never stores energy proportional to its mass in a centralized reservoir. Your fat stores are distributed. Your glycogen reserves are distributed (primarily in liver and muscle). Your ATP pools are distributed across trillions of mitochondria. Evolution ruthlessly eliminated centralized energy storage in mobile organisms because the mass penalty was too severe.
We’ve been fighting evolution’s verdict with brute-force battery engineering. It’s time to stop.
The next generation of humanoid robots will need power architectures as sophisticated as the biological systems they aim to emulate.
Section 5: What Research Exists Today
Already Deployed
Supercapacitors in transportation: Hybrid buses in China use supercapacitor banks for regenerative braking, charging in seconds at stops and discharging during acceleration. Some urban tram systems operate entirely on supercapacitors, recharging at each station. The technology is proven at scale for high-power, frequent-cycle applications.
Grid-scale supercapacitor systems: Power quality applications use supercapacitor arrays to smooth frequency fluctuations and provide instantaneous backup power.
Active Research
Bioinspired Distributed Energy in Robotics: A 2023 paper in Advanced Intelligent Systems directly addressed “Bioinspired Distributed Energy in Robotics and Enabling Technologies,” covering flexible piezoelectric and triboelectric nanogenerators, 3D-printed energy storage, and distributed power architectures for mobile robots[17].
Structural supercapacitors: Research groups at Cambridge, MIT, and Chalmers are developing carbon fiber composites with energy storage capability. Current energy densities remain modest (typically single-digit Wh/kg for the structural component), but the concept is proven and improving rapidly.
Biomorphic batteries: The University of Michigan’s biomorphic battery work demonstrates that when you stop treating the battery as a separate component and instead integrate it into the structural body plan, the energy available to a robot can increase by orders of magnitude[7].
MXene-based energy materials: Doped MXene supercapacitors are being explored as potential “game-changers” for next-generation energy landscapes, with performance characteristics that bridge the gap between batteries and traditional capacitors[18].
The energy storage, power conversion, and motion systems can amount to 30% of the cost of a humanoid robot. (Image: Macquaire) The next frontier: humanoid robots that integrate distributed energy storage into their structural frame, eliminating the centralized battery paradigm entirely.
The Integration Gap
Here’s what’s missing: nobody has assembled these pieces into a complete mitocapacitor architecture for a mobile robot.
We have structural supercapacitors. We have micro-supercapacitors. We have AI-driven energy management. We have bio-electrochemical power generation. But these remain separate research streams. No team has yet built a robot with:
- A carbon fiber composite frame serving as distributed energy storage
- Thousands of embedded micro-supercapacitors positioned at points of actuation
- A slow-rate charging system (whether bioreactor-based or conventional)
- An AI-driven system that predicts motion sequences and pre-charges relevant capacitor clusters
- Fault-tolerant architecture allowing graceful degradation when components fail
This integration challenge is the opportunity.
The Path Forward
In the first two volumes of The Rise of the Machine, I argued that truly capable AI-physical systems would require not just better algorithms but fundamentally rethought embodiment. Better sensors. Better actuators. Better integration between perception and action.
Now I’m arguing for something equally fundamental: better power architecture.
The machines that will move and think like biology will also need to power like biology.
This isn’t just about efficiency, though the efficiency gains would be substantial. It’s about creating robots that can operate in the real world with the grace, adaptability, and endurance of biological systems. A humanoid robot with mitocapacitor architecture could:
- Sustain hours of operation with a fraction of the battery mass
- Handle peak power demands through instantaneous capacitor discharge
- Tolerate component failures through distributed redundancy
- Scale power delivery to match local demand without centralized bottlenecks
- Potentially harvest energy from waste streams, extending operational range indefinitely
The next generation of AI-physical systems, robots that work alongside humans, that move gracefully, that run for hours without returning to charging stations, will be powered by distributed intelligence as much as distributed energy. The control system that predicts which joints will fire in the next 100 milliseconds and pre-charges those specific capacitors is as important as the capacitors themselves.
A Challenge
To the materials scientists: prove that structural supercapacitor composites can achieve energy densities sufficient for practical robotics applications.
To the power electronics engineers: design the DC-DC conversion and distribution systems that can manage thousands of independent energy storage nodes.
To the AI researchers: build the predictive systems that can anticipate power demand milliseconds to seconds in advance.
To the roboticists: stop accepting battery packs as a given and start designing for distributed energy from first principles.
To the bio-electrochemists: push microbial and enzymatic fuel cells toward the power densities that mobile systems demand.
The mitocapacitor is not just a metaphor. It is an engineering specification waiting to be built. Evolution spent two billion years optimizing distributed energy systems for mobile organisms. The designs are proven. The principles are established. The component technologies are emerging.
All that’s missing is the will to assemble them, and the recognition that the battery, as we’ve known it, was always the wrong answer.
The machines are rising. Let’s make sure they can actually keep moving.
References
[1] “Evaluation of potential electric vehicles load-induced damage on flexible pavements.” Transportation Research Part D, 2024. sciencedirect.com
[2] “Hidden effects and externalities of electric vehicles.” Energy Policy, 2024. sciencedirect.com
[3] Giacomello, M. et al. “Molecular mechanisms of mitochondrial dynamics.” Nature Reviews Molecular Cell Biology, 2024. nature.com
[4] “Mechanism of ATP hydrolysis dependent rotation of bacterial ATP synthase.” Nature Communications, 14, 2023. nature.com
[5] Wang, K. et al. “Carbon fiber electrodes for composite structural supercapacitor: Preparation and modification methods.” Journal of Energy Storage, 2024. sciencedirect.com
[6] “Carbon fiber structural battery paves way for light, energy-efficient vehicles.” Chalmers University of Technology, 2024. techxplore.com
[7] Wang, M. et al. “Biomorphic structural batteries for robotics.” Science Robotics, 5(45), eaba1912, 2020. science.org
[8] “Recent developments of advanced micro-supercapacitors: design, fabrication and applications.” npj Flexible Electronics, 4, 2020. nature.com
[9] “Advancements in 2D MXene-based supercapacitor electrodes.” Journal of Materials Chemistry A, RSC, 2024. rsc.org
[10] “Microbial fuel cells powering robots and beyond; historical perspective and future directions.” Biosensors and Bioelectronics, 2025. sciencedirect.com
[11] Wilkinson, S. “‘Gastrobots’: Benefits and Challenges of Microbial Fuel Cells in Food-Powered Robot Applications.” Autonomous Robots, 9, 99–111, 2000. springer.com
[12] Ieropoulos, I. et al. “EcoBot-II: An Artificial Agent with a Natural Metabolism.” International Journal of Advanced Robotic Systems, 2005. sagepub.com
[13] Chakraborty, I. et al. “Glucose-based biofuel cells and their applications in medical implants: A review.” Heliyon, 2024. sciencedirect.com
[14] “Performance of a glucose-reactive enzyme-based biofuel cell system for biomedical applications.” Scientific Reports, 9, 2019. nature.com
[15] “Towards Bio-Hybrid Energy Harvesting in the Real-World.” Applied Sciences, 11(5), 2220, 2021. mdpi.com
[16] “Microbial Batteries Produce Energy from Wastewater and Other Organic Matter.” Stanford Technology Finder. techfinder.stanford.edu
[17] Mukherjee, S. et al. “Bioinspired Distributed Energy in Robotics and Enabling Technologies.” Advanced Intelligent Systems, 2023. wiley.com
[18] “Can Doped-MXene-Based Supercapacitors Be the Game-Changer for Future Energy Landscape?” Langmuir, ACS, 2024. acs.org
[19] Asp, L.E. et al. “A Structural Battery and its Multifunctional Performance.” Advanced Energy and Sustainability Research, 2(3), 2100093, 2021. wiley.com
[20] Gonzalez, C. et al. “Structural composites for multifunctional applications.” Progress in Materials Science, 89, 194–251, 2017. sciencedirect.com
[21] Simon, P. & Gogotsi, Y. “Perspectives for electrochemical capacitors and related devices.” Nature Materials, 19, 1151–1163, 2020. nature.com
[22] Greenman, J. et al. “Microbial fuel cells and their electrified biofilms.” Biofilm, 3, 100057, 2021. sciencedirect.com
[23] Pomerantseva, E. et al. “Energy storage: The future enabled by nanomaterials.” Science, 366(6468), eaan8285, 2019. science.org
[24] Zhu, M. et al. “A Highly Durable All-Polymer Micro-Supercapacitor with Plug-and-Play Function.” Advanced Materials, 33(18), 2006091, 2021. wiley.com
[25] Picard, M. & Shirihai, O.S. “Mitochondrial signal transduction.” Cell Metabolism, 36(6), 1248–1253, 2024. cell.com
Steven Muskal writes at the intersection of molecular biology, materials science, and AI systems. This is apart of “The Rise of the Machine” series. Previous installments explored embodied intelligence and sensorimotor integration.
Steven Muskal, Ph.D. is the CEO of Eidogen-Sertanty, Inc. — a drug discovery informatics company. He has spent four decades working at the intersection of computational biology, AI, and drug discovery. He writes about AI, health, and the intersection of biology and technology at stevenmuskal.com
