Move, don't think: sample-efficient collision avoidance with lossy proximity sensors

Carson Kohlbrenner; Niraj Pudasaini; William Xie; Naren Sivagnanadasan; Nikolaus Correll; Alessandro Roncone

cs.ROOct 22, 2026

Move, don't think: sample-efficient collision avoidance with lossy proximity sensors

Authors: Carson Kohlbrenner, Niraj Pudasaini, William Xie, Naren Sivagnanadasan, Nikolaus Correll, Alessandro Roncone

Organizations: Human Interaction and Robotics Group, University of Colorado Boulder · Correll Lab, University of Colorado Boulder

Abstract

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 $u_{t}$	Description	Distance	Direction	Dimension
$x_{t}$	Raw sensor observation.	–	–	$W\times H\times S$
$f_{\text{tp}}(x_{t})$	Localizes the nearest object within sensing coverage relative to the base of the robot.	$\uparrow$	$\uparrow$	$\downarrow\;(3)$
$f_{\text{min}}(x_{t})$	Returns only the closest range measurement per sensor.	–	$\downarrow$	$\downarrow\;(S)$
$f_{\psi}(x_{t})$	Returns a latent observation.	$\approx$	$\approx$	$\downarrow$
$f_{\text{bin}}(x_{t})$	Returns only binary activations per signal.	$\downarrow$	–	–
$\text{PWM}_{0}(x_{t})$	Returns $[f_{\text{tp}}(x_{t}),f_{\text{tp}}(\hat{x}_{k}),c_{k}]$ .	$\uparrow$	$\uparrow$	$\downarrow(7)$

Table 1: Investigated representations of sensor observation

x

that modify the robot’s access to location-revealing information. We assume the encoder representation

f_{\psi}

performs compression and is well trained, such that

W^{\prime}\times H^{\prime}\times S^{\prime}\times e<W\times H\times 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 $W\times H$ 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 $\hat{x}_{t+1:t+3}$ . b) shows how a rollout will end when contact is predicted, and c) demonstrates the accuracy of the encoder and decoder in $\text{PWM}_{\psi}$ from random samples of a validation set.

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.

Panel of Figure 6 — 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
Sensor Observation	Steps	Speedup	w/ PWM	Steps	Speedup	w/ PWM	NAUC
$f_{\mathrm{tp}}(x_{t})$	338	$8.8\times$ faster	–	427	$>7.0\times$ faster	–	0.508
$\text{PWM}_{0}(x_{t})$	625	$4.7\times$ faster	$1.9\times$ slower	770	$>3.9\times$ faster	$1.8\times$ slower	0.39
$f_{\min}(x_{t})$	770	$3.9\times$ faster	–	1026	$>2.9\times$ faster	–	0.299
$\text{PWM}_{\text{min}}(x_{t})$	1738	$1.7\times$ faster	$2.3\times$ slower	2024	$>1.5\times$ faster	$2.0\times$ slower	0.158
$f_{\psi}(x_{t})$	1613	$1.8\times$ faster	–	2082	$>1.4\times$ 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.

Robot Collision Avoidance

Robot collision avoidance develops sensing, planning, and control methods that keep robots safely separated from people, obstacles, and other robots.

See all 163 papers

Egocentric Tactile and Proximity Sensors as Observation Priors for Humanoid Collision Avoidance
Apr 28, 2026Carson Kohlbrenner, Niraj Pudasaini, William Xie +3RL for Robotics Robotic Tactile Sensing
University of Colorado Boulder, 1111 Engineering Drive, Boulder, CO USA.
Learning Locomotion on Discrete Terrain via Minimal Proximity Sensing
Jun 30, 2026Jiale Fan, Connor Flynn, Tianao Xu +4Terrain-Aware Robot Locomotion Robot Locomotion
All authors are with the Robotic Systems Lab, ETH Zurich.
Whole-Body Planning for Humanoids Navigating Confined Spaces via Self-Collision Avoidance References
Aug 10, 2026Carlos Gonzalez, Luis SentisConstrained Motion Planning Humanoid Robot Locomotion
Department of Aerospace Engineering and Engineering Mechanics, The University of Texas at Austin, TX 78712, USA.

Apr 28, 2026J/K move · Enter open · S save

Egocentric Tactile and Proximity Sensors as Observation Priors for Humanoid Collision Avoidance

Carson Kohlbrenner, Niraj Pudasaini, William Xie, Naren Sivagnanadasan+2

University of Colorado Boulder, 1111 Engineering Drive, Boulder, CO USA.

RL for Robotics Robotic Tactile Sensing Robotic Perception Humanoid Robotics Robot Collision Avoidance

Open paper AskPDF arXiv

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.

Abstract

Robot Collision Avoidance

Explore similar work

Egocentric Tactile and Proximity Sensors as Observation Priors for Humanoid Collision Avoidance

Learning Locomotion on Discrete Terrain via Minimal Proximity Sensing

Whole-Body Planning for Humanoids Navigating Confined Spaces via Self-Collision Avoidance References