Human Inspired Robot Learning to Search Objects in Cluttered Environments Using Maximum Information-based Method
Zhihao Li, Shuaishuai Zheng, Zhipeng Dong, Miao Li, Fei Chen · 2022 12th International Conference on CYBER Technology in Automation, Control, and Intelligent Systems (CYBER) · 2022
Searching for target objects in clutter widely exists in our daily life. However, it is very challenging for robots because of the uncertainty and unobservability caused by cluttered environments. Therefore, this work proposes a method that can imitate the human searching process, spreading the clutter to get more unseen information, which addresses the un-observability problem. This work provides an objective function to conduct the imitation. Due to the uncertainty of cluttered environments, it is hard to obtain environments models. This work uses soft actor-critic, a model-free reinforcement learning method, to address this problem and learn a search policy based on the proposed objective function. We evaluate the training efficiency and search success rate of the policy. The experiments indicate that the proposed objective function can accelerate the training process. The success rates of our method and baseline methods are compared in different difficulty settings. Our method has a higher success rate in highly cluttered environments, including more than 60 objects.