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.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Li, Z., Zhang, Y., Huang, H., Zhou, B., Li, W., Wang, C., Chang, Y., Wei, Y., Zhang, T., Chen, Y. C., Zhou, K., Li, Y., & Cheng, J. (2026). RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning. https://omanscience.com/en/articles/racap-agentic-reasoning-acting-and-coding-as-policies-for-evolvable-robot-learning
MLA 9
Li, Zexi, et al. "RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning." https://omanscience.com/en/articles/racap-agentic-reasoning-acting-and-coding-as-policies-for-evolvable-robot-learning.
Chicago (author–date)
Li, Zexi, Yehang Zhang, Haojian Huang, Bohan Zhou, Wenqian Li, Chenxu Wang, Yifan Chang, Yangkai Wei, Tianyi Zhang, Ying-Cong Chen, Kaiwen Zhou, Yinchuan Li, and James Cheng. 2026. "RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning." https://omanscience.com/en/articles/racap-agentic-reasoning-acting-and-coding-as-policies-for-evolvable-robot-learning.
Harvard
Li, Z., Zhang, Y., Huang, H., Zhou, B., Li, W., Wang, C., Chang, Y., Wei, Y., Zhang, T., Chen, Y. C., Zhou, K., Li, Y. and Cheng, J. (2026) 'RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning', Available at: https://omanscience.com/en/articles/racap-agentic-reasoning-acting-and-coding-as-policies-for-evolvable-robot-learning.
Vancouver
Li Z, Zhang Y, Huang H, Zhou B, Li W, Wang C, et al. RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning. https://omanscience.com/en/articles/racap-agentic-reasoning-acting-and-coding-as-policies-for-evolvable-robot-learning
IEEE
Z. Li, Y. Zhang, H. Huang, B. Zhou, W. Li, C. Wang, Y. Chang, Y. Wei, T. Zhang, Y. C. Chen, K. Zhou, Y. Li, and J. Cheng, "RACaP: Agentic Reasoning, Acting, and Coding as Policies for Evolvable Robot Learning," https://omanscience.com/en/articles/racap-agentic-reasoning-acting-and-coding-as-policies-for-evolvable-robot-learning.