Skip to content

RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning

Sep 2026 · arXiv.org · Vol abs/2609.29394 · 6 citations · ⚡ 1 influential · 25 references
Computer Science

TL;DR

RACaP is introduced, an agentic framework that moves coding to evolution and uses a Reasoning-and-Acting loop to call frozen, typed Policy APIs at deployment and shows that separating reusable code from runtime decisions supports continued evolution, effective transfer, and efficient long-horizon control.

Abstract

General-purpose robot agents must learn from experience, transfer to new tasks, and act efficiently. Code as Policies (CaP) methods generate and repair programs at runtime, incurring latency and entangling reusable mechanisms with task-specific decisions. We introduce RACaP, an agentic framework that moves coding to evolution and uses a Reasoning-and-Acting (ReAct) loop to call frozen, typed Policy APIs at deployment. A two-phase strategy combines capability curriculum learning with autonomous self-evolution to improve the APIs, the ReAct harness, and experience memory. The APIs encode reusable physical mechanisms while exposing arguments for runtime adaptation. ReAct combines task-specific working memory, long-term experience memory, and visual feedback to select actions, verify outcomes, and recover from failures without modifying source code. RACaP achieves 54.4% success on LIBERO-90, 45.0% on zero-shot LIBERO-PRO, and 46.0% on LIBERO-Long, compared with at most 4.0% for CaP baselines on long-horizon tasks. On LIBERO-PRO, it achieves 2.5 times the success rate of CaP baselines and a 1.9-fold speedup in median policy time. For efficient on-robot deployment, rejection-sampled fine-tuning distills GPT-5.6 ReAct decisions into Qwen3-VL-8B-Instruct, yielding a 13.2-fold per-decision inference speedup and reducing repeated physical calls from 16 to 4. These results show that separating reusable code from runtime decisions supports continued evolution, effective transfer, and efficient long-horizon control.

View source

Similar papers

CodeActionBench: Evaluating Agentic Code-as-Policy for Embodied Manipulation

CodeActionBench is introduced, a benchmark of 25 manipulation tasks that evaluates this capability through agentic Code-as-Policy and provides a controlled testbed for measuring how general-purpose models translate their capabilities into manipulation behavior and for examining typical failure scenarios in that process...

Yiheng Lyu, Xueying Jiang, Wen-Hao Li et al. · 3 citations
Sep 2026

Self-Evolving Coding Agents: From Digital Programs to Physical-World Intelligence

Vision-language-action (VLA) and world-action (WAM) models map observations and instructions directly to robot actions. This directness ties a policy to training: minor layout or viewpoint changes cause failure, and instructions generalize poorly. The root cause lies in representation: task requirements, conditions, pr...

Hong-Cheng Gao, Jing-Jing Zhou, Ze Zheng et al. · 7 citations
Preprint Aug 2026

SUN: Agentic Robot Policy Learning with Persistent Task Programs

Model-based control can directly execute specified objectives, while learning can amortize such behaviors into reactive policies, making their combination a natural solution to multi-stage manipulation. We introduce Semantically UNified (SUN) Programs, typed executables that compile grounded relations into aligned opti...

Wei-Qi Wang, Zhi Li, Yuliang Lei et al. · 0 citations

DynaHarness: A Dynamic Physical Harness for Self-Evolving Robot Agents

Pretrained robot policies provide useful action priors, but long-horizon manipulation still requires coordination between semantic reasoning and physical execution. Semantic reasoning operates at a coarser timescale than physical interaction, while episode-level failures provide limited guidance on which system compone...

Hao-Yuan Deng, Jie-Bin Liu, Teng-Xiao Zhang et al. · 2 citations
Sep 2026

ASENA: Self-evolving Agents for Embodied Navigation

ASENA, an embodied agent system that enables general-purpose coding agents to program humanoid navigation, manipulation, and interaction, is presented and ASENA-VLN, a 4B monocular navigation policy that serves as an optional tool within this programmable system, is introduced.

An-Chieh Cheng, Isabella Liu, Edmund Bu et al. · 2 citations
#artificial intelligence Preprint Sep 2026

CAPEX: Efficiently Distilling Foundation Model Behavior into Deployable Robot Policies through Experience-Adaptive Reasoning

CAPEX, an experience-conditioned demonstration collection framework that uses execution experience from previous attempts to adapt how frequently the foundation model must observe, reason, and replan, is introduced and suggests that foundation models can serve as scalable sources of reusable robot experience.

Shivam Aarya, Xi-Jia Zhang, Cheng-Yue Huang et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.