Letting people pick which of several possible robot trajectories they would rather the robot take cut near-collisions by 49.7 percent and raised goal success on a real quadruped
AspireAI community · arXiv
Visuomotor navigation policies have strong perception-action coupling for embodied agents, yet often struggle with safe navigation and dynamic obstacle avoidance in complex real-world environments. CHOP uses counterfactual human preference labels to align visuomotor navigation policies towards human intuition of safety and obstacle avoidance: for each visual observation the robot's executed trajectory is included among a set of counterfactual navigation trajectories, that is, alternatives the robot could have followed under identical conditions, and human annotators provide pairwise preference labels over those trajectories based on anticipated outcomes such as collision risk and path efficiency. The aggregated preferences are then used to fine-tune the policies. On the SCAND dataset, policies fine-tuned with CHOP reduce near-collision events by 49.7%, decrease deviation from human-preferred trajectories by 45.0%, and increase average obstacle clearance by 19.8% on average across multiple state-of-the-art models compared to their pretrained baselines. These improvements transfer to real-world deployments on a Ghost Robotics Vision60 quadruped, where CHOP-aligned policies improve average goal success rates by 24.4% and increase minimum obstacle clearance by 6.8%. The limitation is that the preference labels are collected on one dataset, so the annotators' judgment is baked into the policy and the approach does not remove the need for human annotation, it only changes what is being annotated.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.