Humanoid robots are moving into factories and homes with one unresolved flaw: they run out of energy faster than the shift they are meant to cover. Full-size units typically manage two to four hours, and most packs hold less than 2 kilowatt-hours. A machine costing close to RMB 1 million can still need a socket before the conversation ends.
That gap is pulling automotive suppliers into the category. Joyson Electronics, a tier-one supplier with 25 research centres and 60 plants worldwide, says a solid-liquid hybrid battery developed by its Junen New Energy joint venture has reached mass-production standard and is heading overseas through Joyson’s global channels.
Why a robot battery is not a shrunken car battery
TrendForce data illustrates the constraint. Unitree’s H1 carries 0.864 kWh for under four hours of static runtime. Tesla’s Optimus Gen 2 uses a 2.3 kWh nickel-rich ternary pack for about two hours of dynamic work. Standing upright alone draws 200 to 300 watts because most robot joints have no mechanical self-locking.
Scaling the pack up does not solve it. Torso volume is consumed by joints, motors and controllers, and a heavier pack raises the centre of gravity, which makes dynamic balance harder. Light weight and long runtime pull against each other by design.
The chemistry answer, with numbers
Junen New Energy was founded in May 2026 by Joyson and Beijing Enpower to industrialise solid-state cells. Its first product uses a sulphide solid electrolyte, with a reduced liquid fraction plus a redesigned cell chemistry and pouch structure. Energy density spans 300 to 600 Wh/kg, cell level reaches 380 Wh/kg, cycle life exceeds 2,000, 30 minutes charges to 80 per cent, and combined range improves by about 60 per cent.
Battery management borrows Joyson’s automotive-grade BMS. State-of-charge estimation accuracy reaches 1 to 3 per cent, two to three times better than conventional equipment, and the pack can issue a thermal-runaway warning 30 to 120 minutes ahead. Protection is IP65, with stable operation from minus 20 to 60 degrees Celsius.
Power delivery is the harder requirement. Carrying and flipping demand continuous discharge around 3C, while violent manoeuvres spike to 5C to 15C and, in extreme cases, 20C. Junen offers a full stack of cell, module, pack, BMS, data services and charging, sized for humanoids, wheeled robots, quadruped robots and industrial AGVs.
Standards are the longer game
Junen chief technology officer Xu Wei said the embodied sector still lacks clear safety standards. The company has written an internal standard and is working with lead customers and research institutes on an industry standard, with national-level rules as the next step.
Energy is only one part of Joyson’s robotics plan. Through in-house development plus investment it has added dexterous hands and controllers, with controllers already in production, and is assembling a matrix covering robot brains, AI heads, hands and power.
The European suppliers are running the same play
Joyson is not alone. Since 2025, Bosch, Schaeffler, Valeo and Aptiv have all pushed into embodied intelligence. The logic is technical overlap: perception, domain controllers, algorithms, actuators and thermal management carry over from intelligent vehicles almost unchanged.
Valeo is adapting ultrasonic sensors, cameras, radar, lidar, actuators and power management to robots. Aptiv has launched its PULSE fusion perception system for collaborative and autonomous mobile robots. Schaeffler set up a dedicated embodied robotics company in China in February and plans to deploy thousands of humanoids across its own plants by 2035. Bosch has supplied Germany’s Neura Robotics, set up a robotics company and signed a 2,000-unit order through a Chinese joint venture.
The second advantage is less obvious and harder to copy. Goldman Sachs research found model accuracy dropping from 80 or 90 per cent in simulation to below 50 per cent in real production, where lighting shifts, equipment vibration and material tolerance cannot be simulated. A tier-one supplier with dozens of plants owns exactly the messy environment that robot models need, which turns its own factories into a training asset.
There is also a defensive motive. Price pressure in the car business has flattened growth for many suppliers, and robots offer a second curve with a customer in the same building.
One forecast puts humanoid demand for solid-state batteries above 74 GWh by 2035, more than a thousand-fold increase on 2026. Whether that number holds, the supply chain question has already changed shape: it is no longer who can make a cell, but who owns the production line the robot learns on.
Editor’s note: This is an adapted translation of the original Gasgoo report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://auto.gasgoo.com/news/202610/22I70475591C601.shtml.