Data-driven robots must learn to avoid collisions with as few demonstrations as possible to avoid collecting large datasets of destructive collision examples. We explore whole-body proximity sensors as inductive biases for learning reflexive collision-dodging behavior and find that lossy sensors with low resolution were stronger biases than high-resolution coverage. We also found that, perhaps unintuitively, anticipating when and where contact happens on the body of a robot reduced its learning sample efficiency. We investigated eight candidate representations in a reinforcement learning environment, which included latent compression, future contact anticipation, lossy corruption, and privileged state estimation. The candidate representations were tested by training over 200 humanoid robots to play a simplified version of dodgeball end-to-end under an intentionally constrained sample and compute budget.
Figures & tables
Figure 1: A humanoid robot dodges an incoming projectile using wide coverage, low resolution proximity sensors.
Representation ut
Description
Distance
Direction
Dimension
xt
Raw sensor observation.
–
–
W×H×S
ftp(xt)
Localizes the nearest object within sensing coverage relative to the base of the robot.
↑
↑
↓(3)
fmin(xt)
Returns only the closest range measurement per sensor.
–
↓
↓(S)
fψ(xt)
Returns a latent observation.
≈
≈
↓
fbin(xt)
Returns only binary activations per signal.
↓
–
–
PWM0(xt)
Returns [ftp(xt),ftp(x^k),ck] .
↑
↑
↓(7)
Table 1: Investigated representations of sensor observation x that modify the robot’s access to location-revealing information. We assume the encoder representation fψ performs compression and is well trained, such that W′×H′×S′×e<W×H×S , while retaining relevant information about nearby object distance and direction.
Figure 2: a) The PWM sensor observations are shown here as side-by-side frames of individual sensor depth images, where each sensor occupies an W×H grid of pixels in the observation frame. The green point cloud hits are points on the ball that cross a sensor’s receptive field. When we roll out future observations, we can also visualize them in world space as a point cloud. Here we see the next three predicted observations x^t+1:t+3 . b) shows how a rollout will end when contact is predicted, and c) demonstrates the accuracy of the encoder and decoder in PWMψ from random samples of a validation set.
Figure 3: PWM architecture.
Figure 4: The robot is rewarded for learning collision-avoidant behavior and maintaining posture.
Figure 5: Example of distributed field-sensor activations (left) and ray-sensor activations (right). Lines connect to detected objects.
Figure 6: Sensor representation ablation results. Top: total task success for field and ray sensors across ranges of 0.2, 0.5, 1, and 2 meters. Bottom: task success over training steps for the field and ray sensor representations.
Figure 7: Robots that could anticipate contact (red) consistently underperformed variants that used only current observations (green).
Sensor Observation
10% Success
25% Success
NAUC
Steps
Speedup
w/ PWM
Steps
Speedup
w/ PWM
ftp(xt)
338
8.8× faster
–
427
>7.0× faster
–
0.508
PWM0(xt)
625
4.7× faster
1.9× slower
770
>3.9× faster
1.8× slower
0.39
fmin(xt)
770
3.9× faster
–
1026
>2.9× faster
–
0.299
PWMmin(xt)
1738
1.7× faster
2.3× slower
2024
>1.5× faster
2.0× slower
0.158
fψ(xt)
1613
1.8× faster
–
2082
>1.4× faster
–
0.140
Table 2: Median performance across sensor observation representations. Speedup represents how many less steps were needed to achieve success thresholds of 10% and 25% relative to raw ray sensor observations as the baseline. Observations with the PWM consistently reduced sample efficiency.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 8: Contact prediction accuracy on the validation dataset.
Figure 9: A comparison of Euler step sizes and computation for the dynamics model reveals that only one ODE step is sufficient for reconstruction.
Collision-free motion is often aided by tactile and proximity sensors distributed on the body of the robot due to their resistance to occlusion as opposed to external cameras. However, how to shape the sensor's properties, such as sensing coverage; type; and range, to enable avoidant behavior remains unclear. In this work, we present a reinforcement learning framework for whole-body collision avoidance on a humanoid H1-2 robot and use it to characterize how sensor properties shape learned avoidance behavior. Using dodgeball as a benchmark task, we ablate the properties of sensors distributed across the upper body of the robot and find that raw proximity measurements can substitute for explicit object localization provided the sensing range is sufficient and that sparse non-directional proximity signals outpace dense directional alternatives in sample efficiency.
Carson Kohlbrenner, Niraj Pudasaini, William Xie +3
University of Colorado Boulder, 1111 Engineering Drive, Boulder, CO USA.
Learning-based control has revolutionized dynamic locomotion, yet navigating unstructured terrain remains limited by a robot's incomplete awareness of imminent ground contact. While global perception systems such as LiDARs and depth cameras provide environmental context, they are frequently plagued by latencies, occlusions, and the high computational cost of dense geometric reconstruction. On the other hand, proprioceptive feedback is purely reactive, initiating corrections only after impact has occurred. This work explores embedding a minimal suite of low-cost, high-frequency infrared proximity sensors directly into the feet of a quadrupedal robot. These sensors provide "pre-contact" feedback that is robust to self-occlusions and significantly less computationally demanding than conventional vision-based pipelines. By integrating these localized signals into a reinforcement learning framework, we enable the robot to anticipate terrain discontinuities such as gaps and stepping stones that are problematic for traditional perception stacks due to occlusions or state estimation drift. We demonstrate that such sparse, near-field sensing can be reliably modeled in simulation and transferred to the real world with high fidelity. Experimental results show that local proximity sensing substantially improves traversal robustness over discrete terrain and offers a low-power, low-latency alternative or complement to complex global perception suites in unpredictable environments. For more information about results and methods, please see the project website: https://sites.google.com/view/foot-tof/home.
Jiale Fan, Connor Flynn, Tianao Xu +4
All authors are with the Robotic Systems Lab, ETH Zurich.
Humanoid locomotion in highly confined environments requires navigating dense environmental obstacles and complex self-collision bounds while maintaining multi-contact dynamic feasibility. Traditional trajectory optimizers frequently struggle in these restricted spaces, as navigating the large collision space with splines on particle abstractions is insufficient and leads to poor local minima. To address this, we propose a three-stage whole-body planning framework that formulates kinematic path planning directly over kinematically reachable rigid-body volumes. By integrating differentiable collision avoidance into a reachability-constrained formulation, our framework synthesizes volume-informed guides that reliably guide a full-order trajectory optimizer over long horizons. We show that these optimized plans serve as high-quality references to train a residual reinforcement learning policy for robust online execution. We validate our approach on the Unitree G1 humanoid across three benchmark testbeds exceeding NIST emergency response standards, achieving restricted confinement ratios (Cr<1.5). Our framework generates feasible trajectories across 12-to-18-second tasks with complex foot and hand contacts where standard baselines fail, while the learned policy successfully tracks these plans under extensive domain randomization in physics simulation.
Carlos Gonzalez, Luis Sentis
Department of Aerospace Engineering and Engineering Mechanics, The University of Texas at Austin, TX 78712, USA.