A Two-Stage Trajectory Prediction Algorithm for Mobile Robots Combining the Bayesian and the DMOC Frameworks

Abstract

AbstractIn many real-world robotic applications, predicting an observed agent’s trajectory can significantly improve cooperation efficiency, avoid potential collisions, and alleviate the communication strain in multi-agent scenarios. Given a well predicted trajectory of the observed other agents, the observer can either establish a clear path to its destination or collaborate with the observed agent in a more energy-efficient manner. To address this challenge, in this paper, a two-stage trajectory prediction algorithm is proposed based on the observed agent’s previous trajectory data. First, the potential destination of the observed robot is guessed, and the future guessed path is then sampled using the Monte-Carlo sampling approach within a Bayesian framework. Then, an optimization problem based on a discrete mechanics and optimal control (DMOC) framework with complementarity constraints is proposed to forecast a more reasonable trajectory, while the previously predicted path is used as the reference. Finally, several experiments are undertaken to verify the performance of the proposed algorithm in simulations and real-world applications with our holonomic and nonholonomic mobile robots.KeywordsIntention evaluationMonte-Carlo samplingOptimizationTrajectory predictionComplementarity constraintMobile robotDiscrete mechanics and optimal control