AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics

Xinxin Liu; Bingxin Xu; Wenrui Bao; Ming Li; Wentao Bao; Chen Chen; Yuzhang Shang

cs.ROOct 20, 2026

AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics

Authors: Xinxin Liu, Bingxin Xu, Wenrui Bao, Ming Li, Wentao Bao, Chen Chen, Yuzhang Shang

Organizations: University of Central Florida · University of Southern California · Michigan State University

Abstract

Agentic robotics systems combine planners with robot harnesses to perform complex manipulation tasks. Recent work improves these systems through better harnesses and external memory, but relies on powerful closed-source models as planners and does not improve the planners' own capabilities. In contrast, coding agents have improved substantially by post-training the model itself. We therefore investigate whether post-training can also improve robot planners. However, applying it across robot harnesses requires accommodating different execution mechanisms, including VLA skill invocation and control-code generation. To this end, we introduce AgenticRoboRL, a framework for post-training robot planners across different harnesses. The training recipe combines supervised fine-tuning with multi-turn reinforcement learning. To accommodate different execution mechanisms, harness-specific adapters connect recorded planner interactions with task outcomes, enabling a common RL training pipeline while keeping execution components fixed. On LIBERO-PRO, AgenticRoboRL raises the success rate of a 9B-parameter planner from 18.1% to 71.0% without external memory, approaching the 72.1% achieved by Codex (GPT-5.5) with external memory. These results demonstrate post-training as an effective approach for improving robot planners. We will release the framework and trained planner checkpoints.

Figures & tables

Figure 1: Overview of AgenticRoboRL. (a) Agentic robotics systems combine planners with robot harnesses for multi-turn interaction with the environment. (b) We post-train planners through SFT on successful trajectories followed by multi-turn RL. (c) On LIBERO-PRO, post-training improves a 9B planner’s success rate from 18.1% to 71.0%, approaching the 72.1% achieved by Codex (GPT-5.5)with memory. Mem denotes external memory.

Figure 2: AgenticRoboRL training pipeline. (a) Successful trajectories are converted into training examples for SFT. (b) The planner interacts with the environment through a robot harness, while a proxy records its model calls. (c) Recorded calls are matched with episode outcomes for RL updates, excluding episodes invalidated by infrastructure failures. Harness-specific adapters support a common training pipeline across different harnesses. Only the planner is trained; execution components remain fixed.

Method	Mem.	Spat-T	Spat-S	Obj-T	Obj-S	Goal-T	Goal-S	L10-T	L10-S	Overall
OpenVLA ( Kim et al., 2024 ) †	–	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
$\pi_{0}$ ( Black et al., 2024 ) †	–	0.0	0.0	0.0	2.0	0.0	0.0	0.0	0.0	0.3
$\pi_{0.5}$ ( Intelligence et al., 2025 ) †	–	1.0	20.0	1.0	17.0	2.0	38.0	1.0	8.0	11.0
MolmoAct ( Lee et al., 2026 ) †	–	0.0	0.0	0.0	6.0	0.0	0.0	6.0	0.0	1.5
NORA ( Hung et al., 2025 ) †	–	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
X-VLA ( Zheng et al., 2026 ) †	–	0.0	0.0	8.0	2.0	9.0	1.0	10.0	0.0	3.8

Table 1: Success rate (%) on LIBERO-PRO. Each cell aggregates 100 episodes (10 tasks

\times

10 initial states). T: task perturbation; S: position swap. Rows marked † are taken from Harness VLA ( Zhang et al., 2026c ) ; “/” means not reported. Mem.: the planner retrieves memory at test time. RPent is the harness released with Harness VLA: the Harness VLA rows and the RPent rows use the same harness and the same frozen VLA (

\pi_{\mathrm{RLinf}}

), and differ only in the planner. The row shaded in blue is our final planner. The best entry in each column is in bold.

Method	Camera	Robot	Language	Light	Background	Noise	Layout	Total
OpenVLA ( Kim et al., 2024 ) †	0.8	3.5	23.0	8.1	34.8	15.2	28.5	15.6
OpenVLA-OFT ( Kim et al., 2025 ) †	56.4	31.9	79.5	88.7	93.3	75.8	74.2	69.6
OpenVLA-OFT_w †	10.4	38.7	70.5	76.8	93.6	49.9	69.9	55.8
NORA ( Hung et al., 2025 ) †	2.2	37.0	65.1	45.7	58.6	12.8	62.1	39.0
WorldVLA ( Cen et al., 2025 ) †	0.1	27.9	41.6	43.7	17.1	10.9	38.0	25.0
UniVLA ( Bu et al., 2025 ) †	1.8	46.2	69.6	69.0	81.0	21.2	31.9	43.9

Table 2: Success rate (%) on LIBERO-Plus by perturbation category. Rows marked † are from the LIBERO-Plus leaderboard ( Fei et al., 2025a ) and are single VLA policies; our rows are a planner with the same frozen VLA as in Table 1 (

\pi_{\mathrm{RLinf}}

). The green numbers give the gain over the untrained RPent + 9B row. The row shaded in blue is our final planner. Among methods not trained on LIBERO-Plus, the best entry in each column is in bold.

	Object		Goal		Spatial
Method	Pos	Task	Pos	Task	Pos	Task	Avg.
27B (teacher)	41.0	53.0	12.0	10.0	6.0	0.0	20.3
9B	5.0	5.0	3.0	1.0	1.0	1.0	2.7
9B, SFT (ours)	49.0	39.0	10.0	4.0	10.0	0.0	18.7	+16.0
[0pt][2] 9B, SFT + RL (ours)	67.0	72.0	10.0	16.0	6.0	0.0	28.5	+25.8

Table 3: Success rate (%) with the CaP-X harness on LIBERO-PRO. Pos: position swap; Task: task perturbation. Green: gain over the 9B row. Bold: best per column.

Figure 3: Success on LIBERO-PRO during supervised fine-tuning for the two supervision recipes. Each point is one evaluation on the 800 episodes. The circle marks the checkpoint used as the starting point for RL.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Field	Meaning	RPent	CaP-X
task	Task identifier	Suite and task index	Suite and task index
init	Initial-state identifier; with the task it defines the group	Initial-state seed	Trial index, which fixes the initial state
attempt	Repeat index within the group	Attempt index	Attempt index
reward	Terminal reward	Simulator success check, not the planner’s own finish call	Task-completion flag written by the harness
valid	False for infrastructure failures, which are excluded from optimization	False when the failure is attributed to infrastructure	False for infrastructure failures. Errors in the generated code and unsuccessful execution are valid failures with zero reward

Appendix

Table 5: Fields reported by a harness adapter and their sources in RPent and CaP-X.

Large Language Model-Based Robot Planning

Large language model-based robot planning turns natural-language goals and sensory observations into grounded, executable actions for robots.

See all 235 papers

RHO: Your Coding Agent is Secretly a Roboticist
Jun 15, 2026Karim Elmaaroufi, Justin Svegliato, Sarunas Kalade +3LLM Agent Harnesses Robotic Manipulation
University of California, Berkeley · Embodied Science · AMD
Playful Agentic Robot Learning
Jun 17, 2026Junyi Zhang, Jiaxin Ge, Hanjun Yoo +17Robot Policy Learning Robot Skill Learning
University of California, Berkeley · Impossible Research
Encore: Few-Shot Agentic Discovery of Manipulation Strategies
Sep 29, 2026Yifan Kang, Zihan Wang, Zhiwen Fan +1Robot Policy Learning Language-Conditioned Robot Manipulation
Massachusetts Institute of Technology · AI Open Source Foundation · Texas A&M University +1

Jun 15, 2026J/K move · Enter open · S save

RHO: Your Coding Agent is Secretly a Roboticist

Karim Elmaaroufi, Justin Svegliato, Sarunas Kalade, Graham Schelle+2

University of California, Berkeley · Embodied Science · AMD

LLM Agent Harnesses Robotic Manipulation Robot Policy Learning AI Coding Agents Large Language Model-Based Robot Planning Program Synthesis+4

Open paper AskPDF arXiv

Code-as-Policies (CaP) has shown that large language models (LLMs) can write code to solve robotics tasks by composing perception, planning, and control primitives. Recent CaP systems, however, rely on multi-turn code-generation loops at test time, which is often infeasible for real-time robot control. We introduce Robotics Harness Optimization (RHO), a novel paradigm in which tool-enabled coding agents, at training time, propose and search for interpretable, neurosymbolic multi-file policy repositories (Repositories-as-Policies) that compose these primitives rather than a single prompt, function, or file. RHO searches with reflective feedback from environment reward and execution rather than teleoperation demonstrations. It generalizes to perturbed pick-and-place settings like LIBERO-PRO, where OpenVLA scores 0.0% and

π_{0.5}

averages 12.83%. Using the same low-level primitives, RHO reaches a 45.0% success rate, 2.5x higher than the strongest multi-turn agentic system, and 3.5x higher than

π_{0.5}

. On Robosuite, RHO sets a new state-of-the-art of 70.0%, exceeding the prior multi-turn record of 68.29% using single-turn execution with no corrective LLM code edits at deployment. When an LLM is used in the control loop, as on RAI's O3DE benchmark, RHO optimizes the deployed agent's multi-file harness of prompts, tools, and control code, improving held-out success from 23.5% to 44.3% with 20% less wall-clock time and 27% fewer tool calls.

Abstract

Large Language Model-Based Robot Planning

Explore similar work

RHO: Your Coding Agent is Secretly a Roboticist

Playful Agentic Robot Learning

Encore: Few-Shot Agentic Discovery of Manipulation Strategies