PRTS: A Primitive Reasoning and Tasking System via Contrastive Representations

2026-04-30

AI Summary

This work addresses the limitation of existing vision-language-action models, which reduce pretraining to supervised behavioral cloning and neglect the goal-directed nature and temporal dynamics inherent in robotic learning. The authors propose reframing pretraining through goal-conditioned reinforcement learning, where language instructions are encoded as goals. Within a unified embedding space, contrastive learning aligns state-action embeddings with goal embeddings such that their inner product approximates the probability of goal reachability, thereby implicitly evaluating physical feasibility. Innovatively incorporating goal-reachability awareness enables self-supervised extraction of dense learning signals from offline trajectories without reward annotations. A role-aware causal masking mechanism efficiently integrates this signal into the vision-language backbone. The method achieves state-of-the-art performance on LIBERO, SimplerEnv, and 14 real-world complex tasks, demonstrating substantial improvements in success rate and planning capability—particularly in long-horizon, high-contact, and zero-shot novel instruction scenarios.

Abstract

Vision-Language-Action (VLA) models advance robotic control via strong visual-linguistic priors. However, existing VLAs predominantly frame pretraining as supervised behavior cloning, overlooking the fundamental nature of robot learning as a goal-reaching process that requires understanding temporal task progress. We present \textbf{PRTS} (\textbf{P}rimitive \textbf{R}easoning and \textbf{T}asking \textbf{S}ystem), a VLA foundation model that reformulates pretraining through Goal-Conditioned Reinforcement Learning. By treating language instructions as goals and employing contrastive reinforcement learning, PRTS learns a unified embedding space where the inner product of state-action and goal embeddings approximates the log-discounted goal occupancy, the probability of reaching the language-specified goal from the current state-action, quantitatively assessing physical feasibility beyond static semantic matching. PRTS draws this dense goal-reachability supervision directly from offline trajectories without reward annotations, and folds it into the VLM backbone via a role-aware causal mask, incurring negligible overhead over vanilla behavior cloning. This paradigm endows the high-level reasoning system with intrinsic goal reachability awareness, bridging semantic reasoning and temporal task progress, and further benefits goal-conditioned action prediction. Pretrained on 167B tokens of diverse manipulation and embodied-reasoning data, PRTS reaches state-of-the-art performance on LIBERO, LIBERO-Pro, LIBERO-Plus, SimplerEnv, and a real-world suite of 14 complex tasks, with particularly substantial gains on long-horizon, contact-rich, and zero-shot novel-instruction settings, confirming that injecting goal-reachability awareness significantly improves both execution success and long-horizon planning of general-purpose robotic foundation policies.

Problem

• Vision-Language-Action models
• goal-reaching
• temporal task progress
• robotic control
• goal reachability

Innovation

• Goal-Conditioned Reinforcement Learning
• Contrastive Representations
• Vision-Language-Action Models
• Goal Reachability Awareness
• Offline Robot Learning