ABot-Claw: A Foundation for Persistent, Cooperative, and Self-Evolving Robotic Agents

Dongjie Huo; Haoyun Liu; Guoqing Liu; Dekang Qi; Zhiming Sun; Maoguo Gao; Jianxin He; Yandan Yang

cs.CVApr 11, 2026

ABot-Claw: A Foundation for Persistent, Cooperative, and Self-Evolving Robotic Agents

Authors: Dongjie Huo, Haoyun Liu, Guoqing Liu, Dekang Qi, Zhiming Sun, Maoguo Gao, Jianxin He, Yandan Yang,

Organizations: AMAP CV Lab Alibaba Group

Abstract

Current embodied intelligent systems still face a substantial gap between high-level reasoning and low-level physical execution in open-world environments. Although Vision-Language-Action (VLA) models provide strong perception and intuitive responses, their open-loop nature limits long-horizon performance. Agents incorporating System 2 cognitive mechanisms improve planning, but usually operate in closed sandboxes with predefined toolkits and limited real-system control. OpenClaw provides a localized runtime with full system privileges, but lacks the embodied control architecture required for long-duration, multi-robot execution. We therefore propose ABot-Claw, an embodied extension of OpenClaw that integrates: 1) a unified embodiment interface with capability-driven scheduling for heterogeneous robot coordination; 2) a visual-centric cross-embodiment multimodal memory for persistent context retention and grounded retrieval; and 3) a critic-based closed-loop feedback mechanism with a generalist reward model for online progress evaluation, local correction, and replanning. With a decoupled architecture spanning the OpenClaw layer, shared service layer, and robot embodiment layer, ABot-Claw enables real-world interaction, closes the loop from natural language intent to physical action, and supports progressively self-evolving robotic agents in open, dynamic environments.

Figures & tables

Figure 1: System Architecture of ABot-Claw. Featuring a Layered Decoupling Strategy Among Interaction, Shared Services, and Embodiment Execution.

Figure 2: Overview of the ABot-Claw architecture. Building upon the OpenClaw runtime, ABot-Claw enables persistent, cooperative, and self-evolving robotic agents through three core components: (1) a unified embodiment interface for coordinating diverse robot platforms, (2) a visual-centric multimodal memory for long-term, cross-embodiment context retention, and (3) a critic-based closed-loop feedback mechanism that supports online progress evaluation, local refinement, and dynamic replanning. Together, these components close the loop between high-level intent and low-level action in open, dynamic environments.

Figure 3: Visual-centric multi-robot memory. ABot-Claw maintains a unified memory service that stores object memory, place memory, visual memory, and supports object history search, spatial radius search, text-based or image-based retrieval, and image-based retrieval.

Figure 4: Workflow of OpenClaw Layer . After receiving a user instruction, OpenClaw first loads the available skills. The robot layer then acquires robot-specific information and invokes service-layer modules, such as memory and process supervision. Based on the current observations, the user request is decomposed into a sequence of subtasks, from which executable Python code is generated and dispatched to the robot. During execution, OpenClaw continuously monitors robot states and task progress, enabling timely correction of execution errors.

Figure 5: Demonstration of grasping an object outside the manipulator’s initial field of view. The bottle is not observed at the beginning. After a clarification query to the user, the manipulator searches the tabletop and successfully grasps the target object. Processes 1 to 5 represent the robotic arm’s search process; the target object was found in process 6.

Figure 6: Demonstration of semantic manipulation under ambiguous instructions. Given an abstract user request, the manipulator interprets the semantic intent, selects the appropriate object, and completes the grasp-and-place task.

Figure 7: Demonstration of humanoid mobile manipulation for delivery. Given the instruction to deliver the coffee bottle in front of the robot to the person in Room 207, the humanoid grasps the object, navigates to the target room, and hands it to the recipient.

Figure 8: Demonstration of humanoid-assisted inspection. When the user requests the status of the quadruped robot, the system first detects that the quadruped is disconnected. It then queries the memory module to retrieve the robot’s last known location and dispatches the humanoid robot to inspect its condition.

Figure 9: Demonstration of quadruped-guided visitor reception. Given the instruction to meet a visitor at the elevator and guide the visitor to the meeting room, the system selects the quadruped robot, dispatches it to the elevator, detects the visitor, and leads the visitor to the destination.

Self-Evolving Agents

Self-evolving agents improve their skills, memory, tools, policies, or environments by learning from interaction and evaluation feedback.

See all 398 papers

ROSClaw: A Hierarchical Semantic-Physical Framework for Heterogeneous Multi-Agent Collaboration
Apr 6, 2026Rongfeng Zhao, Xuanhao Zhang, Zhaochen Guo +4VLMs for Robotics Cross-Embodiment Robot Learning
Shanghai Research Institute for Intelligent Autonomous Systems, Shanghai 201210, China · National Key Laboratory of Autonomous Intelligent Unmanned Systems (Tongji University), Shanghai 201210, China · Frontiers Science Center for Intelligent Autonomous Systems, Shanghai 201210, China +1
ROSClaw: An OpenClaw ROS 2 Framework for Agentic Robot Control and Interaction
Mar 27, 2026Irvin Steve Cardenas, Marcus Anthony Arnett, Natalie Catherine Yeo +2Robotics Robot Foundation Models
Advanced Telerobotics Research Lab, Kent State University, Kent, OH, USA · OpenDive Technologies, New York City, NY, USA · PlaiPin Inc, Singapore
RoboClaw: An Agentic Framework for Scalable Long-Horizon Robotic Tasks
Mar 12, 2026Ruiying Li, Yunlang Zhou, YuYao Zhu +15Long-Horizon Robotic Manipulation VLMs for Robotics
AgiBot, China · National University of Singapore · Shanghai Jiao Tong University, Shanghai 200240, China +1

Apr 6, 2026J/K move · Enter open · S save

ROSClaw: A Hierarchical Semantic-Physical Framework for Heterogeneous Multi-Agent Collaboration

Rongfeng Zhao, Xuanhao Zhang, Zhaochen Guo, Xiang Shao+3

Shanghai Research Institute for Intelligent Autonomous Systems, Shanghai 201210, China · National Key Laboratory of Autonomous Intelligent Unmanned Systems (Tongji University), Shanghai 201210, China · Frontiers Science Center for Intelligent Autonomous Systems, Shanghai 201210, China+1

VLMs for Robotics Cross-Embodiment Robot Learning Multi-Robot Coordination Long-Horizon Robotic Manipulation Continual Robot Learning

Open paper AskPDF arXiv

The integration of large language models (LLMs) with embodied agents has improved high-level reasoning capabilities; however, a critical gap remains between semantic understanding and physical execution. While vision-language-action (VLA) and vision-language-navigation (VLN) systems enable robots to perform manipulation and navigation tasks from natural language instructions, they still struggle with long-horizon sequential and temporally structured tasks. Existing frameworks typically adopt modular pipelines for data collection, skill training, and policy deployment, resulting in high costs in experimental validation and policy optimization. To address these limitations, we propose ROSClaw, an agent framework for heterogeneous robots that integrates policy learning and task execution within a unified vision-language model (VLM) controller. The framework leverages e-URDF representations of heterogeneous robots as physical constraints to construct a sim-to-real topological mapping, enabling real-time access to the physical states of both simulated and real-world agents. We further incorporate a data collection and state accumulation mechanism that stores robot states, multimodal observations, and execution trajectories during real-world execution, enabling subsequent iterative policy optimization. During deployment, a unified agent maintains semantic continuity between reasoning and execution, and dynamically assigns task-specific control to different agents, thereby improving robustness in multi-policy execution. By establishing an autonomous closed-loop framework, ROSClaw minimizes the reliance on robot-specific development workflows. The framework supports hardware-level validation, automated generation of SDK-level control programs, and tool-based execution, enabling rapid cross-platform transfer and continual improvement of robotic skills. Ours project page: https://www.rosclaw.io/.

Abstract

Self-Evolving Agents

Explore similar work

ROSClaw: A Hierarchical Semantic-Physical Framework for Heterogeneous Multi-Agent Collaboration

ROSClaw: An OpenClaw ROS 2 Framework for Agentic Robot Control and Interaction

RoboClaw: An Agentic Framework for Scalable Long-Horizon Robotic Tasks