Learning to Compose: Recursive Self-Improvement of Shared Executable Stacks for Zero-Shot Long-Horizon Robot Manipulation

Yangzhi Chen; Feiran Liu; Zishuo Wang; Zijian Jin; Xin Zhu; Lecheng Ruan; Wenjuan Han

cs.AIOct 17, 2026

Learning to Compose: Recursive Self-Improvement of Shared Executable Stacks for Zero-Shot Long-Horizon Robot Manipulation

Authors: Yangzhi Chen, Feiran Liu, Zishuo Wang, Zijian Jin, Xin Zhu, Lecheng Ruan, Wenjuan Han

Organizations: Beijing Jiaotong University · New York University · University of Southern California · PKU-HEC joint lab for industrial intelligence, Peking University ·

Abstract

Individually competent robot skills can fail when composed: tools may lack a required capability, completion signals may misrepresent physical progress, and one skill may leave a state that violates its successor's prerequisites. We introduce Recursia, a recursive self-improvement framework that treats the shared executable stack, rather than individual task programs, as the persistent unit of improvement. The stack integrates tools, skills with execution contracts, handoffs, and verifiers. Source-side execution traces guide failure attribution and the generation of missing tools or repairs to existing shared components. Generality checks, mechanism tests, and paired selection and regression gates determine which changes become the next parent. The resulting stack is frozen and reused through declarative task adapters, allowing source-derived changes to support new compositions without target-side code repair. Experiments cover LIBERO-Pro Long simulation and a physical Franka FR3 equipped with a Franka Hand. In simulation, Recursia improves success over ASPIRE by 5.8 percentage points under Pos perturbations and 15.1 points under Task perturbations. Across three physical compositional target tasks, it improves mean success over the Skill Document Only baseline by 40 percentage points. Analyses of constituent execution, candidate rejection, and inference drivers distinguish local mechanism improvements from complete task success and characterize the cost of feedback-conditioned orchestration. These results examine shared executable interfaces as a concrete mechanism for compositional transfer in robot programming.

Figures & tables

Figure 1: Framework overview. The improvement loop operates on four executable levels. Both newly generated tools and repairs to existing components must pass the same acceptance process before entering the shared stack. Target execution does not modify the stack.

Figure 2: Representative shared-stack interfaces.

Figure 3: Zero-shot transfer to LIBERO-Pro Long. (a) Position perturbations ( Pos ). (b) Task-specification perturbations ( Task ). Baseline success rates are reproduced from the existing comparison; Recursia achieves success rates of 0.284 on Pos and 0.534 on Task . Aggregate and per-task tables appear in Appendix D .

Task	SD-only	Recursia
T1: Shared basket	0.0	20.0
T2: Reassigned plates	60.0	80.0
T3: Bowl–mug dependency	20.0	100.0
Average	26.7	66.7

Table 1: Success rates (%) on physical compositional target tasks using Franka FR3 with Franka Hand.

Figure 4: Recorded compositional generalization cases. Original RGB recordings and official predicate timelines for the N=6 mug–pudding sequence. (a) Seed 42 completes the mug goal, but pudding binding returns not_found ; the earlier goal remains true through termination. (b) Seed 41 completes both goals while preserving the first. Gray segments denote false predicates; blue and teal segments denote true predicates. Across ten swap-suite episodes, the mug predicate is true at termination in all ten, whereas the full conjunction succeeds in one. The two illustrated seeds have different initial states. RGB frame identifiers and simulation-step timestamps use separate counters.

Figure 5: Inference-driver comparison on a frozen stack of the $N=13$ iteration. (a) Pos and (b) Task success rates, each 50 matched episodes. (c) Mean end-to-end episode time, including program generation and robot execution, for all episodes.

Figure 6: Recorded validation-gate cases. (a) Initial state. Original simulator frames from the N=8 iteration, both with seed 86. (b) Development and selection successes for this task. (c) The iteration N=9 reduced a local failure mechanism but failed the predeclared outcome gains. (d) The iteration N=11 improved the aggregate selection score while losing two previously successful basket-placement states, exceeding the per-task loss cap.

Entry	Failure	Repair	Recorded Result
N=10	Lack of front/middle reference relations	Shared instance selection and relation dispatch	Dispatch $81/90\!\rightarrow\!84/90$ ; gains measured on source tasks.
N=11	Rotary localization, pinch, and arc defects	Shared turn-on/off driver	Turn-on $10/10$ in each panel.

Table 2: Representative repairs. Rates refer to the specified tasks, not the complete Long-Pro benchmark.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Query	Required fields	Allowed values
Point localization	abstain , x , y , confidence , reason	Boolean abstention; integer pixel coordinates; confidence: low , medium , or high .
Mask selection	choice , reason	A candidate image label, or none .
Identity / mask review	verdict , reason	yes , no , or unsure .

Appendix

Table 3: Structured outputs for the inference prompts. Every response includes a brief reason.

Figure 7: Recursia transfer scaling with source-task experience. Frozen-stack success rates on Pos and Task are shown at the supplied iteration checkpoints.

Robotic Manipulation

Robotic manipulation studies how robots perceive, plan, and control physical interactions to grasp, move, assemble, and use objects.

See all 1,685 papers

RoboRSI: Stable, efficient, and reusable robot self-evolution in complex real-world environments
Oct 8, 2026Zimo Wen, Yijin Chen, Yuxuan Cao +8Robot Skill Learning Continual Robot Learning
Exploring the Limits of Combinatorial Generalization in Embodied Manipulation
Oct 11, 2026Xupeng Zhang, Yantai Yang, Chang Guo +2Generalization in Robotic Manipulation Long-Horizon Robotic Manipulation
AutoLab, School of Artificial Intelligence, Shanghai Jiao Tong University
ASPIRE: Agentic /Skills Discovery for Robotics
Jun 30, 2026Runyu Lu, Yubo Wu, Ethan Kou +13Robot Policy Learning Long-Horizon Robotic Manipulation
NVIDIA · UMich · UIUC +2

Oct 8, 2026J/K move · Enter open · S save

RoboRSI: Stable, efficient, and reusable robot self-evolution in complex real-world environments

Zimo Wen, Yijin Chen, Yuxuan Cao, Wendi Chen+7

Robot Skill Learning Continual Robot Learning Robotic Manipulation

Open paper AskPDF arXiv

A generalist robot should not only perform diverse tasks but also improve through experience, turning what it learns during execution into capabilities that later tasks can reuse. Robot agents that act through code can already repair programs from execution feedback, yet it remains a central challenge to organize this experience around the task structure that gives it meaning, so that each repair is attributed to the responsible capability, supported by execution evidence, and validated before it is reused. We introduce RoboRSI, a robot self-improvement system built on Top-Down Skill Refinement (TSR). TSR decomposes tasks into compound, atomic, and base skills with scoped responsibilities and explicit input--output contracts, attributes each execution outcome to the responsible branch, and confines revision to that branch. Building upon this structure, a Manager, Planner, Engineer, and Reviewer coordinate planning, execution, diagnosis, and the validated release of new skills, while people steer the process through objectives and corrections; stable skill sequences are further consolidated into reusable compound skills. On a mobile manipulator, RoboRSI develops multi-object household cleanup over 104 rounds. In simulation, it achieves the highest success rate on LIBERO, LIBERO-PRO, LIBERO-Plus, and RoboTwin, exceeding the strongest baseline by 2.7 to 11.0 percentage points.

Abstract

Robotic Manipulation

Explore similar work

RoboRSI: Stable, efficient, and reusable robot self-evolution in complex real-world environments

Exploring the Limits of Combinatorial Generalization in Embodied Manipulation

ASPIRE: Agentic /Skills Discovery for Robotics