Phase-Conditioned Affordance Fields for SE(3)-Equivariant Visuomotor Policies

Key points

• However, existing SE(3)-equivariant visuomotor policies build this bias on a global equivariance assumption that treats the entire observation as governed by a single rigid transformation, which breaks down in real-world manipulation tasks where the task-relevant region changes across different objects.1
• We address this by learning a phase-conditioned affordance field (PAF) over the observed point cloud, a per-point score that the SE(3)-equivariant policy uses in two complementary ways.1
• This introduces a local geometric inductive bias, while preserving the policy's global SE(3) equivariance.1
• We validate on multiple MimicGen tasks under varying spatial randomization, and PAF consistently improves the success rate of the underlying policy.1

Sources (1)

• [1]Phase-Conditioned Affordance Fields for SE(3)-Equivariant Visuomotor Policies
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 06:35 PM

  SE(3)-equivariance provides a useful geometric inductive bias for visuomotor policies, encoding that actions covary with rigid transformations of the observation.

  However, existing SE(3)-equivariant visuomotor policies build this bias on a global equivariance assumption that treats the entire observation as governed by a single rigid transformation, which breaks down in real-world manipulation tasks where the task-relevant region changes across different objects.

Extractive summary: sentences quoted from the sources.