🤖 Humanoid 🦾 Industrial & Cobot 🚚 AGV / AMR 🐕 Quadruped ⚙️ Reducers · Servos · Sensors 🚁 Drones & Autonomy 🧠 Embodied AI
Robos News
Robotics

Direction-Conditioned Policies via Compositional Subgoal Scoring for Online Goal-Conditioned Reinforcement Learning

arXiv:2606.16515v1 Announce Type: cross Abstract: Hamilton-Jacobi-Bellman theory implies that the optimal goal-conditioned action depends on the goal only through the gradient of the goal-reaching distance at the current state, yet standard online GCRL still conditions the actor on the raw goal -- a signal that is geometrically uninformative when the goal is far from the data distribution. We propose Direction-Conditioned Policies (DCP), a fully online method that decomposes goal-reaching into

Direction-Conditioned Policies via Compositional Subgoal Scoring for Online Goal-Conditioned Reinforcement Learning

Published June 16, 2026 · Category: Robotics

Overview

arXiv:2606.16515v1 Announce Type: cross Abstract: Hamilton-Jacobi-Bellman theory implies that the optimal goal-conditioned action depends on the goal only through the gradient of the goal-reaching distance at the current state, yet standard online GCRL still conditions the actor on the raw goal -- a signal that is geometrically uninformative when the goal is far from the data distribution. We propose Direction-Conditioned Policies (DCP), a fully online method that decomposes goal-reaching into two components sharing one InfoNCE representation $\psi$: a subgoal-scoring step that selects a visited state $z_t$ aligned with the final goal $g$ in $\psi_g$, and a direction-conditioned actor that consumes the unit direction $d_t$ and magnitude $r_t$ from $\psi(s_t)$ to $\psi(z_t)$. The two components train jointly, factor cleanly at deployment (subgoal scoring is removed, while direction conditioning remains with $g$ in place of $z_t$), and admit independent modification at the same $(d_t,r_t)$ interface. We prove three results. First, direction sufficiency under HJB: the optimal action under control-affine dynamics depends on the goal only through the value gradient. Second, a quantitative bound showing that, under mild conditions on the learned representation and assuming the scoring rule returns an on-path $z_t$, the actor's conditioning input at training and at deployment coincide up to representation error and geodesic slack. Third, a controllable-subspace characterization of when directional conditioning fails. Across nine environments, DCP improves over Contrastive RL on most final metrics, with the largest gains on manipulation and obstacle-interaction tasks; a qualitative analysis of the learned $\psi$-distance landscape shows the contrastive representation behaves as an online quasimetric encoding environment topology, and the single failure case (AntSoccer) localizes to a learned-gradient pathology that the theory anticipates.

Source

Originally published at arxiv.org.

Related Articles

CD
Robos News Newsroom

Robos News covers markets, crypto and commodities for Asia & the Middle East — tier-1 desk research, AI-driven analysis, institutional-grade data. Tip our newsroom: [email protected]

Email the newsroom →
Disclaimer: This article is for informational purposes only and does not constitute investment advice. Data may be delayed up to 15 minutes. Past performance is not indicative of future results. Consult a licensed financial advisor before making investment decisions.

Related Stories

More from News →