Proprioceptive Sketches as Long-Horizon Intent for Generative Action Policies

2026-10-01

AI Summary

This study addresses the lack of explicit long-horizon intent in generative robot policies and the prohibitive cost of existing solutions by proposing a proprioceptive action model. The method employs a single Transformer denoiser with block-causal attention within a diffusion framework to jointly generate compact, temporally unaligned joint-space path sketches and dense executable action chunks, thereby guiding action generation via geometric intent. Temporal invariance is achieved through arc-length parameterization, while interleaved denoizing preserves the directional dependence of actions on sketches. Extensive evaluations demonstrate that this approach significantly improves success rates by up to 75% in the Push-T simulation and four real-world bimanual tasks, validating its effectiveness.

Abstract

Generative robot policies predict short action chunks but lack explicit long-horizon intent. Recent methods expose longer-horizon structure through language plans, subgoal images, or video forecasts, which are costly to generate and still need to be translated into robot motion. Predicting future robot motions avoids this translation, but a dense, time-indexed trajectory requires numerous parameters to cover the full remaining task, and over a short horizon it largely repeats the action chunk and adds little guidance for action generation. We propose Proprioceptive Action Models (PAM), which jointly generate a compact, timing-free sketch of the robot's remaining joint-space path and a dense executable action chunk within a single transformer denoiser. The sketch parameterizes the path by arc length rather than time, capturing geometric intent invariant to execution timing. Block-causal attention and a staggered denoising schedule maintain directed sketch-to-action dependence, ensuring the action tokens condition on a progressively cleaner sketch throughout sampling. In simulation, PAM improves over its action-only counterparts on Push-T and LIBERO-Long; on four real-world bimanual tasks, it raises success from 47.5% to 75.0%. Project page: https://nicehiro.github.io/pam_dp/

Problem

• generative robot policies
• long-horizon intent
• action generation
• robot motion prediction

Innovation

• Proprioceptive Action Models
• Generative robot policies
• Arc-length parameterization
• Block-causal attention
• Staggered denoising