HEATS: A Hierarchical Framework for Efficient Autonomous Target Search with Mobile Manipulators

TL;DR

HEATS addresses autonomous target search in unknown, cluttered environments by coupling a mobile-manipulator–specific target viewpoint planner with a whole-body motion planner. It introduces a viewpoint state library and a dual-stage decision module to achieve comprehensive exploration and inspection while minimizing detours, supported by a global path search (Hybrid A* and ST-RRT*) and a real-time IPC optimization using Relaxed Barrier Functions within DDP. The approach yields improved search completeness, reduced path length, and lower joint movement costs in both simulated and real-world tests, using onboard sensing and computation. The work is open-sourced to benefit the community and lays groundwork for extending perception with Vision-Language Models to further bias search toward high-probability target regions.

Abstract

Utilizing robots for autonomous target search in complex and unknown environments can greatly improve the efficiency of search and rescue missions. However, existing methods have shown inadequate performance due to hardware platform limitations, inefficient viewpoint selection strategies, and conservative motion planning. In this work, we propose HEATS, which enhances the search capability of mobile manipulators in complex and unknown environments. We design a target viewpoint planner tailored to the strengths of mobile manipulators, ensuring efficient and comprehensive viewpoint planning. Supported by this, a whole-body motion planner integrates global path search with local IPC optimization, enabling the mobile manipulator to safely and agilely visit target viewpoints, significantly improving search performance. We present extensive simulated and real-world tests, in which our method demonstrates reduced search time, higher target search completeness, and lower movement cost compared to classic and state-of-the-art approaches. Our method will be open-sourced for community benefit.

Figures (9)

• Figure 1: The mobile manipulator searches for targets (represented by apriltags) in both real indoor environments and simulation environments. The target search completeness is enhanced by leveraging the high degree of freedom of the mobile manipulator, allowing the camera at the end-effector to cover all areas of the scene.
• Figure 2: The overview of the proposed hierarchical framework for efficient autonomous target search.
• Figure 3: The mobile manipulator detects frontiers in simulation. (a) Apriltags are randomly placed within an office environment and used as search targets. (b) The robot updates the map information using LiDAR and depth data while inspecting $V_{occ}$. Based on the properties of the voxels, exploration frontiers and inspection frontiers are extracted and clustered.
• Figure 4: Generating viewpoint state library for a frontier cluster. The yellow arrows represent uniformly sampled mobile base poses, combined with arm states to generate whole-body state of the mobile manipulator. $\varDelta R$ and $\varDelta \theta$ represent the radial resolution and angular resolution during sampling, respectively. Each whole-body state corresponds to a viewpoint.
• Figure 5: Results of the dual-stage decision module. (a) The region-level decision-making determines the visitation order of regions based on the partitioning results (black arrows and numbers). (b) The viewpoint-level decision-making determines the visitation order of viewpoints within the current region and proceeds to the target region (blue arrows).