A Novel Predictive Safety Criteria for Robust Collision Avoidance of Autonomous Robots

Abstract

In this article, we propose a novel predictive strategy for collision avoidance of multiple mobile robots by precisely quantifying the robot’s impending collisions. Our proposed approach is based on formulating the collisions by predicting the robots’ states and provides generalized safety conditions that are more realistic measures for collisions. This is a novel safety condition that extends the previous methods that use distance or velocity for safety conditions into a more generalized format to effectively avoid underestimating or overestimating the collisions by explicitly considering surrounding objects’ motions. We formulate the safety condition based on the robots’ motion into a vector format and construct a predictive safety matrix. We then consolidate the safety condition based on the proposed safety matrix’s eigenvalues. To effectively incorporate the safety condition into a control design considering uncertainties in the input sensor measurements, we perform uncertainty analysis on our proposed safety condition, the results of which are further used in this article to design a robust collision avoidance and tracking controller. This tracking controller avoids collisions effectively by incorporating the proposed safety condition into a robust tracking control design that is mathematically proven to be stable and guaranteeing safe motion. We show that the proposed safety condition, by effectively quantifying the conflicts among the robots, avoids collisions while achieving improved tracking performance compared to other existing reactive methods. This method is also verified experimentally and is shown to be effective in real-time control with superior tracking performance to the existing methods.