On legged robots, sustained locomotion often ends in thermal shutdown rather than battery depletion: the heat of walking concentrates in a few joint motors, which overheat long before the battery is empty. We train thermally aware locomotion policies by reinforcement learning without ever observing temperature, and demonstrate this on a Unitree Go2 quadruped. We first identify a coupled 13-node lumped RC thermal network, fitted on two runs of a single robot and calibrated against the measured electrical power, that predicts all twelve motor temperatures with a mean RMSE below 1 ∘C across different robot instances, gaits, payloads, and speeds. Under a sustained gait, the temperatures the motors will settle at follow directly from the gait-averaged squared torques and joint speeds. Our core contribution is a reward that drives down the hottest predicted asymptotes: thermal time constants exceed the credit-assignment horizon of reinforcement learning by orders of magnitude, but predicted asymptotes respond to torque and joint speed within a single gait cycle, so the policy learns to lower and spread heat with no temperature input. On the robot, the policy turns sustained locomotion from temperature-limited to battery-limited: with a 5 kg payload at 1 m/s, a learned baseline reaches thermal shutdown after 19 minutes (1.1 km), whereas our policy runs the battery down after 174 minutes (10.0 km) with its hottest motor 14 ∘C below the shutdown threshold in a 26 ∘C room; on an unloaded factory circuit with stairs and ramps, the baseline overheats after 25 laps (6.2 km), while our policy completes 47 (11.2 km) on one battery with 22 ∘C of margin.
Figures & tables
Fig. 1: Sustained 1 m/s walking with a 5 kg payload. The baseline reaches thermal shutdown after 19 min (top); our thermal policy drains the battery after 174 min with 14 °C of headroom (middle). Grey: the 12 measured motor temperatures; shaded: min–max of our thermal model, projected with constant gait inputs; dots: predicted asymptotes (blue hip, orange thigh, green calf), those above the axis omitted. Bottom: the payload and its mounting.
Fig. 2: The two training runs, following protocols A and B. The twelve temperatures (light-grey solid lines) are recorded every two seconds. Segments are designed and arranged to produce a large range of temperatures while keeping their maximum (dark-grey solid line) under Go2’s thermal shutdown threshold of about 86 °C.
Model
k
Train
Val
Test
all
hot
all
hot
all
hot
22-parameter model (ours)
22
0.78
0.74
0.88
0.85
0.97
0.85
1: no electrical identification
22
0.77
0.71
0.90
0.88
1.04
0.98
2: no joint-velocity heating
21
0.79
0.76
1.00
1.07
1.25
1.56
3: no bypass share to body
21
0.95
1.15
1.35
1.66
1.83
2.57
4: no quiescent power terms
20
1.14
1.09
1.68
2.02
1.87
1.82
TABLE I: Performance of the thermal model and its ablations: RMSE in °C of the displayed motor temperatures over all 12 joints (all) and of the maximum over the twelve joints (hot), averaged over runs. Train: the two training runs (fit); val: the nine validation runs; test: twelve held-out runs scored once. k counts all identified parameters (electrical + thermal). Ablations 2–5 are the best structure of their family selected by LORO.
Fig. 3: Compactness frontier over 1,129 structural variants of the network, scored on the validation runs. Red: best structure per parameter count by validation RMSE; blue: by LORO. Numbers mark ablations 1–5 from Tab. I ; arrows point off-scale.
Fig. 4: Hottest predicted asymptote versus payload at 1 m/s. Lines: simulation, means and sample s.d. across five seeds × five checkpoints. Crosses: the two deployed runs of our policy, wth=−0.2 , evaluated at the simulation ambient of 30 °C.
Fig. 5: Tracking errors among 5 s survivors relative to the baseline (1; lower is better). Means and sample s.d. across five seeds × five checkpoints, after averaging payloads, are divided by the baseline mean. Grey bands show baseline s.d.; numbers below each row give its absolute mean.
Fig. 6: Views of the factory circuit, showing the gait of our thermal policy across different terrain types.
Fig. 7: Factory circuit, no payload: hottest joint of the baseline (purple) and our thermal policy (green); measured (solid) and predicted (dashed).
Motor thermal management is often overlooked in the context of electrically-actuated robots, particularly legged robots, but motor overheating is a key factor that limits long-duration locomotion especially under payload conditions. This paper integrates a whole-body thermal model of a quadruped robot into the reinforcement learning pipeline to update motor temperatures, and proposes a two-stage training framework for motor thermal management. In this framework, a nominal policy is first pre-trained as a locomotion baseline capable of traversing diverse terrains. A residual policy is then trained on top of the nominal policy to provide corrective actions based on the robot's thermal state, ensuring high performance under low-temperature conditions and preventing motor overheating under high-temperature conditions. Simulation results demonstrate that the proposed policy achieves an effective balance between motor thermal safety and locomotion performance. Real-world experiments on a Unitree A1 quadruped robot further validate the approach: under a 3 kg payload, the robot achieves stable locomotion across multiple terrains for over 13 minutes, while the nominal policy alone leads to motor overheating in about 5 minutes.
Yuhang Wan, Weixian Lin, Letian Qian +5
School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan, 430074, China
Electrically-actuated quadrupedal robots possess high mobility on complex terrains, but their motors tend to accumulate heat under high-torque cyclic loads, potentially triggering overheat protection and limiting long-duration tasks. This work proposes a thermal-aware control method that incorporates motor temperatures into reinforcement learning locomotion policies and introduces thermal-constraint rewards to prevent temperature exceedance. Real-world experiments on the Unitree A1 demonstrate that, under a fixed 3 kg payload, the baseline policy triggers overheat protection and stops within approximately 7 minutes, whereas the proposed method can operate continuously for over 27 minutes without thermal interruptions while maintaining comparable command-tracking performance, thereby enhancing sustainable operational capability.
Letian Qian, Yuhang Wan, Shuhan Wang +1
State Key Laboratory of Intelligent Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, China
Learning-based quadrupedal locomotion typically relies on complex reward formulations that entangle task specification, operational limits, gait preference, and terrain adaptation within a single optimization objective. We instead treat these functions through distinct mechanisms: rewards for task specification, constraints for operational limits, energy minimization for gait preference, and exteroceptive perception for adapting energy use to terrain difficulty. We show that these components jointly enable efficient, terrain-adaptive locomotion, and that removing each component exposes a distinct failure mode. Our formulation removes explicit gait priors (including air-time, contact-count, and foot-clearance targets) in favor of emergent behavior. Compared to a conventional complex-reward baseline, our formulation achieves comparable terrain traversal while reducing cost of transport by 56% and operational-limit violations by 96%. The resulting policies transfer zero-shot to a physical Unitree Go2 using LiDAR-based elevation mapping. Project website with videos: https://tinyurl.com/locomposition.
Loukas Kordos, Leonard T. Franz, Simon Rappenecker +4
Technical University of Munich. · University of Tübingen. · Hertie Institute for Clinical Brain Research & Center for Integrative Neuroscience. +1