Geometry-aware safety-critical local reactive controller for robot navigation in unknown and cluttered environments

Project Description

Driven by the ever-increasing need for robots to interact seamlessly with their surrounding environments, the topic of safety in motion has emerged as a critical aspect of robotic autonomy. The planning and executing of a feasible robot trajectory often rely on complete knowledge of the environment, but there could be situations where the environment is partial or fully unknown. Hence, it is desirable to design a highly reactive local planner to facilitate swift and accurate responses to unforeseen environmental changes and ensure collision-free manoeuvres. Our work develops a mathematical approach to compute the real-time optimal controls for robots. We formulate an optimisation problem to minimise the violence to the local path, while using positivity of some polynomial functions to describe the collision-avoidance safety constraints. The optimisation problem is solved by the semidefinite relaxation method based on Sum-of-Squares and Moment problems. Moreover, we implement this novel algorithm on real robots where local waypoints in the unknown and cluttered environment are extracted from sensors, and the robot's real-time movement is given by the optimal control obtained from our algorithm. Both numerical simulations and real-world implementation show our approach enables better passability for complex environments with unknown obstacles and guarantees collision avoidance.

Project Investigator

Dr TANG Xindong  (Department of Mathematics)

Project Collaborators

• Hong Kong University of Science and Technology
• The Hong Kong University of Science and Technology (Guangzhou)

Data Analytics and Artificial Intelligence

EXPLORE MORE