cs.RONov 10, 2026

Camera equalisation of an 11.5x drop in environmental illuminance: four imitation-learning policies on an SO-101 arm

Authors: Simon Cowling

Organizations: fysical.ai, Odense, Denmark

Abstract

Real-robot benchmarks of learned visuomotor policies often record room illuminance as the lighting condition. We show that, on a typical low-cost rig, room light and image light can be almost unrelated. Four imitation-learning policies (ACT, SmolVLA, Diffusion Policy and pi0.5) were trained on the same 50 demonstrations of a pick-and-place task on an SO-101 arm. Each was run for ten rollouts at about 520 lux and again at about 45 lux, an 11.5x reduction. Success was 20/40 in normal light and 19/40 in low light: a change of -2.5 percentage points (95% CI -23 to +19), with no evidence that the policies differed in sensitivity (interaction p = 0.86). The recorded video shows why: mean image brightness was only 3% lower in low light on the front camera and 19% lower on the wrist camera. Both cameras were set to automatic exposure and white balance, which most likely absorbed the difference; exposure was not locked in this experiment. An accidental repeat of one 10-rollout block reproduced its success count (7/10) but matched the first block at only 6 of 10 placements, consistent with chance. We conclude that lux alone does not describe what a policy sees. Camera settings, and image statistics compared with the training recordings, are the quantities to report and control. Data and code are public.

Figures & tables

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Oct 27, 2026

From 12 to 200 Demonstrations: How ACT, Diffusion Policy, SmolVLA and π0.5π_{0.5} Scale on a Low-Cost Arm

We trained four widely used imitation-learning policies, ACT, Diffusion Policy, SmolVLA and π0.5π_{0.5}, on nested subsets of 12, 25, 50, 100 and 200 teleoperated demonstrations of one pick-and-place task with distractor objects, using one fixed training recipe per policy. Every trained model was then run on a physical SO-101 arm on a fixed set of object placements: 1,340 scored episodes in 126 blocks. ACT, trained from scratch, rose from 1 success in 40 episodes at 12 demonstrations to 37 in 40 at 100. π0.5π_{0.5} with its action expert fine-tuned was the best policy when data was scarce (16 of 40 at 12 demonstrations, 28 of 40 at 25) and then stayed near 70%; from 100 demonstrations ACT was ahead, by a margin this study does not fully resolve. SmolVLA and Diffusion Policy peaked near 50%, and both were affected by a failure to start: SmolVLA's arm never started moving in 47% of episodes, and Diffusion Policy hardly ever started until an idle lead-in was trimmed from the demonstrations. For π0.5π_{0.5} the fine-tuning recipe mattered more than the amount of data: low-rank adaptation succeeded in 4 of 110 episodes, against 130 of 200 for action-expert fine-tuning. From the joint and camera recordings we place every episode on a four-stage ladder (started, reached the ball, carried it, delivered it) and show that for ACT more data moves failures from reaching, to grasping, to none, while the failures π0.5π_{0.5} retains are in grasping. We also measure how weak a ten-episode block is as a measurement (the same checkpoint tested twice differs by 2 successes in 10 on average), separate seed-to-seed variation from that noise, and audit the human scoring against the recordings. Training length was fixed within each recipe, so the amount of data and the number of times it was seen change together; we state which conclusions survive that confound and which do not.
Jun 7, 2026

Benchmarking Vision-Language-Action Models on SO-101: Failure and Recovery Analysis

Vision-Language-Action (VLA) models have demonstrated strong generalization in robotic manipulation, yet existing evaluations are primarily conducted in simulation or on expensive robotic platforms, leaving their robustness on affordable real-world robots largely unexplored. We present a standardized real-world benchmark for evaluating representative VLA and imitation learning policies on the low-cost SO-101 robotic platform. The benchmark comprises four representative manipulation tasks together with unified evaluation protocols, enabling systematic comparison under embodiment uncertainty. Using real-world teleoperated demonstrations, we fine-tune and evaluate π0.5π_{0.5}, SmolVLA, Wall-X, and ACT directly on the physical platform. Beyond conventional task success rates, the benchmark incorporates a structured failure taxonomy, semantic- and execution-level failure decomposition, and recovery-aware evaluation metrics to characterize policy robustness. Experimental results show that stronger pretrained VLA policies generally outperform the imitation learning baseline, although performance remains highly task-dependent under low-cost robotic deployment conditions. Execution instability emerges as the dominant failure source, while recovery capability varies substantially across architectures. These results highlight the importance of failure and recovery analysis beyond binary task success and establish SO-101 as a practical benchmark for evaluating embodied AI systems under realistic low-cost robotic deployment conditions.
Sep 26, 2026

An Empirical Study on What Matters for Viewpoint-Generalizable Policies in Visual Imitation Learning

Visual imitation learning is a promising approach to training robot manipulation policies capable of completing a wide variety of tasks. However, policies today remain brittle to viewpoint perturbations, making deployment in diverse environments a challenge. We present a controlled empirical study of which design choices allow visuomotor policies to generalize across viewpoints. We find that viewpoint generalization improves when dense visual tokens are retained and the action head participates in geometric reasoning. On a suite of simulated tasks that span a wide range of camera poses, we show that these design choices yield a policy that remains performant across viewpoints. As a practical consequence, a policy trained with these design choices also transfers zero-shot from simulation to the real world under random camera configurations.