Harmonized Learning with Concurrent Arbitration: A Brain-inspired Motion Planning Approach

Tianyuan Jia, Chaoqiong Fan, Qixin Wang, Han Yuyang, Xia Wu

Primary Area: applications to robotics, autonomy, planning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Brain-inspired learning;Motion planning;Deep reinforcement learning;

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We propose a harmonized learning with concurrent arbitration inspired by concurrent reasoning of the prefrontal cortex for motion planning in high-dimensional continuous environments.

Abstract: Motion planning, regarded as a sequential decision-making problem, poses a challenge for robots in high-dimensional continuous environments due to inefficient sampling. In contrast, humans inherently possess a distinctive advantage in decision-making by leveraging limited information, primarily relying on the concurrent reasoning mechanism in the prefrontal cortex. Motivated by this, we propose a brain-inspired Deep Reinforcement Learning scheme for planning, called Harmonized Learning with Concurrent Arbitration (HLCA). The approach effectively mimics human capacity for concurrent inference tracks and the ability to harmonize strategies. Specifically, in the planning process, a general Concurrent Arbitration Module (CAM) is meticulously crafted to balance the exploration-exploitation dilemma simply and efficiently. Besides, the harmonized style facilitates robots self-improving learning during the learning process, enabling the selection of appropriate strategies to guide planning. Experimental results show that HLCA outperforms the state-of-the-art benchmarks in terms of three representative metrics, which confirms the potential of emulating human-like capabilities to enhance the intelligence and efficiency of robotic planning.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1290