Energy efficiency in Physical AI depends on two converging fronts: better battery chemistry and smarter drive system architecture, both currently in active development.
Physical AI robots face a dual energy problem: batteries that degrade unpredictably and drive systems that waste power through inefficient conversion.
If you want to understand why humanoid robots still have limited runtime, you need to look at two separate but connected systems. First, the energy source: the battery. Second, the energy consumer: the actuator and its drive electronics. Both are under active engineering pressure right now, and recent developments from MIT and NORD Drivesystems show just how much headroom is still left on the table. From a builder perspective, the interesting thing is that these two problem spaces rarely get discussed together, even though they are deeply interdependent.
MIT engineers discovered that dendrites, the metal filaments that cause battery short circuits, actually grow faster under low mechanical stress, reversing what most researchers assumed.
The conventional assumption in solid-state battery research was straightforward: apply more pressure to the solid electrolyte and dendrite growth would be suppressed. More compression, fewer failure points. MIT's new research flips that logic. According to Interesting Engineering, the MIT team found that dendrites grow faster when mechanical stress is lower, not higher. This is a significant finding because it means that previous design choices aimed at reducing stress may have been actively accelerating the failure mode engineers were trying to prevent. Here is what the data shows: the relationship between mechanical stress and dendrite propagation is not linear and not intuitive. The failure mechanism is more nuanced than the field had assumed.
Dendrites are tiny metal filaments that can grow through a battery's electrolyte layer over time, eventually bridging the two electrodes and causing a short circuit. In liquid electrolyte batteries, this is a known failure mode. In solid-state batteries, the expectation was that the rigid solid electrolyte would physically block dendrite propagation. MIT's finding shows that is not guaranteed, and the failure conditions are different from what the field modeled. For robotics applications where batteries go through thousands of charge cycles and operate under variable mechanical loads, understanding the real failure conditions matters a great deal.
The MIT research is both a setback and a step forward. It is a setback in the sense that it reveals an underappreciated failure mode. It is a step forward because identifying the real mechanism is the prerequisite to solving it. According to Interesting Engineering, the finding challenges old theories, which implies previous engineering decisions made under those theories may need revision. For Physical AI hardware designers counting on solid-state batteries to deliver higher energy density and better safety profiles, this kind of foundational research is the work that actually determines whether next-generation batteries arrive on time.
Modular drive systems like those shown by NORD Drivesystems at MODEX allow engineers to configure motors and drives precisely for each application, reducing energy waste from over-engineered or mismatched components.
On the actuator side of the energy equation, NORD Drivesystems is demonstrating a different approach to efficiency at the MODEX automation event. According to The Robot Report, NORD's modular product portfolio offers high configurability for compact, efficient, and reliable drive systems engineered to individual application needs. The key word here is configurability. One of the persistent inefficiencies in industrial and robotic drive systems is the mismatch between what a motor is capable of and what an application actually demands. When you run an oversized motor at partial load, you lose efficiency. When you run an undersized motor near its thermal limit, you lose reliability. Modular architecture is an attempt to close that gap.
NORD's framing of their systems as supporting intelligent automation points to something worth unpacking. Drive intelligence means the system can adapt its output to real-time demand rather than running at fixed parameters. According to The Robot Report, NORD's systems are designed for intelligent automation, which at the drive level typically means variable frequency control, real-time diagnostics, and the ability to integrate with higher-level automation logic. For humanoid and mobile robots, this kind of adaptive drive behavior is important because load conditions change constantly depending on posture, terrain, and task.
Better batteries and more efficient drives address different parts of the same problem. Improving one without the other leaves significant efficiency gains unrealized.
Let me break down the components here. A robot's energy budget has two sides: how much energy it can store, and how efficiently it converts that stored energy into mechanical work. Solid-state batteries, once their dendrite failure modes are understood and controlled, promise higher energy density and better thermal stability than current lithium-ion packs. That means more runtime or smaller, lighter packs for the same runtime. But if the drive system wastes a significant fraction of that energy through poor matching, thermal losses, or inefficient conversion, the gains from better batteries are partially cancelled out. The honest trade-off is this: chemistry improvements are slow and expensive. Drive system optimization is faster and more immediately actionable. Both matter.
Humanoid robots face the most demanding energy constraints in Physical AI: variable loads, tight weight budgets, and long operation cycles that stress both batteries and drive systems simultaneously.
Humanoid robots are not a forgiving application. Unlike a fixed industrial arm that operates in predictable cycles, a humanoid robot changes its load profile constantly. Walking, lifting, balancing, reaching, all create different torque and power demands across dozens of joints. This variability is exactly where both the MIT battery research and the NORD modular drive approach become relevant. A battery that degrades unpredictably because its failure mechanism was misunderstood is a serious problem in a robot that needs to operate reliably over thousands of hours. A drive system that wastes energy through poor application matching compounds the battery problem by shortening runtime and increasing thermal load. According to The Robot Report, NORD's systems target compact, efficient, and reliable performance, which maps directly to the constraints humanoid designers face.
These two developments point in the same direction: assumptions embedded in component design need to be tested, and configurability matters more than raw specification numbers.
Here is what stands out when you look at both stories together. MIT's battery finding is a reminder that widely held engineering assumptions can be wrong, and that the failure mode you think you are solving may not be the one that is actually occurring. NORD's modular drive approach is a reminder that matching components to real application demands beats specifying for peak theoretical performance. Both insights are relevant to anyone designing Physical AI hardware. The energy chain from battery to actuator to mechanical output is full of assumptions that deserve scrutiny. The field is still early, and foundational research like MIT's is precisely the kind of work that resets what builders think they know.
MIT engineers found that dendrites, the metal filaments that cause short circuits in solid-state batteries, grow faster under low mechanical stress rather than high stress. This reverses a widely held assumption and suggests previous design strategies may have been accelerating the failure mode they were trying to prevent.
Modular drive systems allow engineers to configure motors and drive electronics precisely for the demands of a specific application. This reduces energy wasted by oversized or mismatched components running outside their optimal operating range, which is a common source of inefficiency in robotic systems.
Not yet at commercial scale. MIT's 2026 research reveals that dendrite failure mechanisms in solid-state batteries are more complex than previously modeled. Solving these failure modes is a prerequisite to reliable deployment. The research represents progress, but also confirms how much foundational work remains.
Humanoid and mobile robots operate under constantly changing load conditions. A drive system optimized for one condition runs inefficiently under others. High configurability, as demonstrated by NORD Drivesystems at MODEX, allows the system to match its output to real demand rather than running at fixed parameters regardless of actual load.
They are two parts of the same energy chain. Better batteries provide more stored energy, but drive system efficiency determines how much of that energy converts into useful mechanical work. Improving one without the other leaves significant gains unrealized. Both need to advance together for meaningful runtime improvements in Physical AI systems.