Receding Horizon Chance Constraint Motion Planning for Uncertain Multi-Agent

Link to the paper

Robotics Research Assistant:

Thesis: On Receding Horizon Chance Constraint Motion Planning for Uncertain Multi-Agent Systems

• Developed a cooperative, multi-agent path-planning algorithm for an uncertain and dynamic environment
• Implemented the algorithm in a Python environment to check for feasibility and effectiveness
• Used mathematical tools like Chance Constraint, theory of stochastic systems, Gaussian noise, prediction update through Kalman filter, etc to formulate the algorithm

Abstract:

This work presents a coordinated algorithm for a multi-agent autonomous robot system to navigate through an uncertain and dynamic environment. Using the concepts of receding horizon chance constraint we've developed a path-planning algorithm for the multi-agent systems. This algorithm deals with the problem of motion planning in an unbounded uncertain environment by providing a trade-off between the risk of collision and the infeasibility of the path. Using the sampling-based rapidly-exploring random trees (RRT) planner, we've incorporated for uncertainty within the formulation. Sensor measurements are used to improve our belief in the localization of the robot and to decrease the chance of blowing up the covariance in a multi-agent system. The planning is decentralized and a coordination strategy is developed to account for constraints that are transmitted between the agents. By defining linear constraints and with the assumption of Gaussian noise the feasibility of the trajectory is established. Simulations are performed to verify the feasibility and the algorithm generated collision-free paths by generating trajectories around other agents and obstacles in multi-agent scenarios with dynamic obstacles.

GitHub Repository

Algorithm at Work :