Robots can expand their capabilities by using tools and tailor their choices to the particular circumstances. However, current robotics are still a long way from creating strategies using situational tools.
A recent article on arXiv.org proposes an integrated learning and planning framework in which robots understand and devise effective tool use strategies by inferring about the physical properties of the device. contribute to the success of the mission.
The robot Kuka is quietly writing a version of Martin Luther’s bible. Image credit: Marc Wathieu via WikimediaCC-BY-2.0
The framework focuses on the physical effects produced by the engine and learns to recognize the essential physical properties to accomplish the task. Then different strategies for using the tool whose effectiveness is measured by joint effort are created.
The researchers demonstrate the framework’s ability to perform two robotic tasks: breaking walnuts and cutting carrots. It can identify physical attributes critical to mission success and create an effective tooling strategy using visible and invisible objects as tools.
We present a robotic learning and planning framework that produces an effective tooling strategy with the least amount of overhead, capable of handling subjects that differ from training. Leverage a finite element method (FEM)-based simulator that reproduces detailed, continuous visual and physical effects with observed instrument-use events, essential physical properties contribute to the effects identified through the proposed Symbolic Regression (IDSR) algorithm. We continue to develop an optimal control-based motion planning scheme to integrate the kinematics and dynamics of the robot and the specific tool to create an efficient trajectory that exhibits the characteristics of the robot. have learned. In simulation, we demonstrate that the proposed framework can generate more effective tool-using strategies, which are radically different from those observed in the two sample tasks.
Research articles: Zhang, Z., Jiao, Z., Wang, W., Zhu, Y., Zhu, S.-C. and Liu, H., “Understanding physical effects for effective tool use”, 2022. Links: https://arxiv.org/abs/2206.14998
