LOGIV: Graph-Verified Execution with Causal Local Repair for Long-Horizon Robot Tasks

Enqiao Lu; Yuxiang Xie; Xingrui Yu; Yuyang Wang; Yipeng Gu; Zhenglin Wan; Pengfei Zhou; Wangbo Zhao

cs.ROOct 29, 2026

LOGIV: Graph-Verified Execution with Causal Local Repair for Long-Horizon Robot Tasks

Authors: Enqiao Lu, Yuxiang Xie, Xingrui Yu, Yuyang Wang, Yipeng Gu, Zhenglin Wan, Pengfei Zhou, Wangbo Zhao,

Organizations: CUHKSZ · A*STAR · NTU · COCO Matrix ·

Abstract

Agentic robots must turn a high-level goal into actions while keeping the physical conditions needed for future steps valid. A completed action does not ensure that its effects will persist: a later disturbance can invalidate support that the remaining plan still assumes. We present LOGIV, a graph engineering framework for agentic robot execution that uses an explicit task graph to connect symbolic planning, observation-based verification, and embodied action. LOGIV compiles a validated plan into condition-support and resource-ordering edges, checks their physical validity during execution, and uses dependency tracing to determine what must be restored after a failure. Repair is constrained by retained unfinished actions, and the combined continuation is symbolically validated before execution resumes. The framework operates above existing Vision-Language-Action (VLA) and World Action Model (WAM) policies without adding trainable components. Across 20 dual-arm tasks from RoboTwin~V2 and BiCoord with three execution backbones, LOGIV improves mean task success without injected disturbances by 10.67 and 15.33 percentage points, respectively. Under predefined perturbations, it achieves higher task completion and condition restoration than three adapted recovery baselines for every backbone on both benchmarks. On three real-robot tasks, successful trials increase from 12/30 to 17/30. These results support explicit, continuously checked task dependencies as a practical interface for reliable agentic robot execution.

Figures & tables

Fig. 1: LOGIV execution on the BiCoord build_bridge task. (A) A validated reference plan is compiled into the initial task DAG before execution. (B) A previously placed support is displaced, invalidating a condition required by the remaining plan. (C) LOGIV repairs only the affected support operations, retains the unaffected beam operations, and resumes execution on the updated task DAG.

Fig. 2: LOGIV pre-execution planning and task-graph construction. The Operator Planner generates the reference plan $P$ , where each $o_{i}$ in panel (B) is a grounded high-level PDDL operator. After validation by VAL, $P$ is compiled into the initial task graph $G$ .

Fig. 3: LOGIV online verification and topology-guided causal local repair. The three gates establish reliable task facts, verify the selected subtask, and check graph dependencies. Upon a failure, the Repair Gate identifies affected nodes, validates a local repair plan, rebuilds the task graph, and resumes execution. Human assistance is reserved for future evaluation.

Backbone	Method	RoboTwin V2			BiCoord
Backbone	Method	C $\uparrow$	P $\uparrow$	R $\uparrow$	C $\uparrow$	P $\uparrow$	R $\uparrow$
$\pi_{0.5}$	Backbone	56	32	18	39	31	21
	VeriGraph-PDDL	68	56	66	46	39	56
	LERa-PDDL	67	57	64	47	43	63
	VLM-RP-BT	63	58	68	45	45	59
	LOGIV	72	60	72	50	48	68

TABLE I: Task completion and condition restoration across backbones and online correction methods.

Panel of Figure 5 — Fig. 4: Real-robot rollouts of the three tasks. Top: pick the tomato and place it in the bowl. Middle: build the block structure with both arms. Bottom: plug in the Ethernet cable.

Task	Platform	$\pi_{0.5}$	$\pi_{0.5}$ + LOGIV
pick_tomato	Astribot S1	7 / 10	9 / 10
build_blocks	Astribot S1	3 / 10	5 / 10
insert_cable	AgileX Piper	2 / 10	3 / 10
Overall	—	12 / 30 (40.0%)	17 / 30 (56.7%)

TABLE II: Real-robot success over ten trials per task.

Fig. 5: Ablation results of LOGIV. The top panel shows task completion and condition restoration under perturbations. The middle and bottom panels show modified nodes and repeated valid actions during repair, respectively. Full: complete method; -State: no State Gate; -UNRES: no uncertainty handling; -Node: no Node Gate; -Graph: no Graph Gate; -Causal: no causal dependency tracing; -Local: replanning all remaining actions instead of local repair; -VAL: no logical validation of the repaired plan.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

RoboTwin V2 Chen et al. (2025a)
beat_block_hammer	blocks_ranking_size
handover_block	move_can_pot
open_microwave	place_dual_shoes
stack_blocks_three	stack_bowls_three
stamp_seal	turn_switch
BiCoord Peng et al. (2026b)

Appendix

TABLE A1: The 20 dual-arm tasks from RoboTwin V2 and BiCoord.

Agentic AI for Robotics

See all 46 papers

Learning Robust Execution in Robotic Manipulation with Agentic Reinforcement Learning
Jul 15, 2026Xiaopeng Zhang, Yueyang Weng, Qi Liu +2Agentic RL Robot Failure Recovery
School of Inteligence Science and Engineering, the Harbin Institute of Technology Shenzhen, 518055, China. · Faculty of Robot Science and Engineering, Northeastern University, Shenyang, 110819, China.
GAVEL: Graph World Models for Verified and Efficient Long-Horizon LLM Task Planning
Sep 16, 2026Ruiyang Wang, Hao-Lun Hsu, Swarajh Mehta +3Belief-Space Planning Large Language Model-Based Robot Planning
Refuse the Harm, Complete the Task via Graph-of-Rules Planning for LLM-Enabled Robots
Oct 17, 2026Ali Taheri, Alireza Taban, Qizhou Wang +1LLM Agent Safety Large Language Model-Based Robot Planning
Nanyang Technological University · Queen’s University · RIKEN AIP

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

Learning Robust Execution in Robotic Manipulation with Agentic Reinforcement Learning

Xiaopeng Zhang, Yueyang Weng, Qi Liu, Yongjin Mu+1

School of Inteligence Science and Engineering, the Harbin Institute of Technology Shenzhen, 518055, China. · Faculty of Robot Science and Engineering, Northeastern University, Shenyang, 110819, China.

Agentic RL Robot Failure Recovery Robotic Manipulation Robust RL Robustness of VLA Models

Open paper AskPDF arXiv

Robotic manipulation poses fundamental challenges due to uncertainty, long-horizon execution, and compounding errors, which can easily destabilize execution and lead to task failure. Although recent vision-language-action (VLA) models exhibit strong generalization, they typically lack explicit mechanisms to assess execution stability and to recover when execution deviates from its nominal behavior. In this paper, we propose: (1) two complementary metrics to assess execution quality at runtime, and (2) an agentic reinforcement learning framework that learns to restore effective execution through high-level decision-making rather than directly learning low-level actions. In this framework, an agentic policy reasons over recent execution history and selects among a small set of execution modes to regulate the execution process. Under execution degradation, it triggers appropriate recovery mechanisms to restore the robot to previously visited nominal states, enabling the task to continue. We evaluate the proposed method on the LIBERO benchmark, achieving up to a 13.7% improvement in success rate under standard settings and up to a 39.2% improvement under disturbance settings, demonstrating substantially enhanced execution robustness.

Abstract

Agentic AI for Robotics

Explore similar work

Learning Robust Execution in Robotic Manipulation with Agentic Reinforcement Learning

GAVEL: Graph World Models for Verified and Efficient Long-Horizon LLM Task Planning

Refuse the Harm, Complete the Task via Graph-of-Rules Planning for LLM-Enabled Robots