A real-time hierarchical control method for safe human–robot coexistence

In modern industrial environments, robots are expected to work close to human operators collaborating with them in completing various tasks, e.g., collaborative assembly or delivery of objects. However, it is difficult to take task optimization into account under the premise of ensuring safety in a human-involved dynamic environment. Therefore, a real-time hierarchical control method containing two hierarchical optimal controllers with complementary functions is proposed to tackle the above problems. The upper-layer Model Predictive Controller aims at performing primary tasks, such as end-effector pose tracking, singularity, and joint limit avoidance. It is formulated as a Bayesian Inference problem with a Gaussian process prior and an exponential likelihood function. The resulting maximum a posteriori estimation problem can be solved efficiently using the Matrix-Scaled Stein Variational Gradient Descent and GPU. The upper-layer optimal controller aims at performing primary tasks, such as end-effector pose tracking, singularity, and joint limit avoidance. The lower-layer safety-critical controller, formulated as a constrained quadratic programming problem, is responsible for tracking the output interpolation of the higher-layer controller while respecting the safety constraints constructed in the form of Stochastic Control Barrier Functions. Both of the optimal controllers run repeatedly but with different frequencies (upper-layer controller: 20Hz, lower-layer controller: 40Hz). The proposed method provides a solution to deal with both collision avoidance and task constraints. Lastly, we conducted real-time simulation and experimental tests on a Doosan robot, and the test results validated the effectiveness of the proposed method.