Human-Robot Collaborative Tele-Grasping in Clutter With Five-Fingered Robotic Hands

Teleoperation offers the possibility of enabling robots to replace humans in operating within hazardous environments. While it provides greater adaptability to unstructured settings than full autonomy, it also imposes significant burdens on human operators, leading to operational errors. To address this challenge, shared control, a key aspect of human-robot collaboration methods, has emerged as a promising alternative. By integrating direct teleoperation with autonomous control, shared control ensures both efficiency and stability. In this letter, we introduce a shared control framework for human-robot collaborative tele-grasping in clutter with five-fingered robotic hands. During teleoperation, the operator's intent to reach the target object is detected in real-time. Upon successful detection, continuous and smooth grasping plans are generated, allowing the robot to seamlessly take over control and achieve natural, collision-free grasping. We validate the proposed framework through fundamental component analysis and experiments on real-world platforms, demonstrating the superior performance of this framework in reducing operator workload and enabling effective grasping in clutter.

I. Introduction

In Hazardous environments and catastrophic scenarios, such as search and rescue operations, disaster response efforts, or nuclear facilities, the utilization of robots to substitute human operators becomes imperative to mitigate safety risks. While fully autonomous robots hold the potential to perform intricate operational tasks, they often encounter challenges related to instability and unreliability when operating in dynamic, unpredictable, and unstructured environments. Teleoperation emerges as a viable solution to circumvent the limitations of full autonomy, combining the decision-making abilities of human intelligence with the robust manipulation capabilities of robots. It offers advantages such as rapid decision-making, flexibility, reliability, and strong adaptability. However, it heavily relies on the operator's experience, resulting in lower operational efficiency and weak repeatability. This reliance can burden operators and lead to errors, exacerbated by the inability of operators to access complete real-time information from remote environments.