Safety vs. Efficiency: AI-Based Risk Mitigation in Collaborative Robotics

The use of AI-based risk mitigation is increasing to provide safety in the areas of smart manufacturing, automated logistics etc, where the human-robot collaboration operations are in use. This paper presents our work on implementation of fuzzy logic system (FLS) and reinforcement learning (RL) to build risk mitigation modules for human-robot collaboration scenarios. Risk mitigation using FLS strategy is developed by manually defining the linguistic values, tuning the membership functions and generating the rules based on ISO/TS15066:2016. RL-based risk mitigation modules are developed using three different Qnetworks to estimate the Q-value function. Our purpose is twofold: to perform a comparative analysis of FLS and RL in terms of safety perspectives and further to evaluate the efficiency to accomplish the task. Our results present that all the proposed risk mitigation strategies improve the safety aspect by up to 26% as compared to a default setup where the robot is just relying on a navigation module without risk mitigation. The efficiency of using FLS model is maintained to the default setup, while the efficiency of using RL model is reduced by 26% from the default setup. We also compare the computation performance of risk mitigation between centralized and edge execution where the edge execution is 27.5 times faster than the centralized one.