Path planning with moving obstacles using stochastic optimal control
This work addresses collision-free navigation for robots in human-populated environments, representing an incremental improvement over existing methods.
The paper tackles the problem of robot path planning in dynamic environments with moving obstacles by formulating it as a stochastic optimal control problem and introducing a tractable approximation solved via an online rollout framework. Simulation results show the method achieves nearly the same expected time to target as receding horizon A* while maintaining a larger expected minimum distance to obstacles, and outperforms CBF-based methods when higher obstacle clearance is desired.
Navigating a collision-free and optimal trajectory for a robot is a challenging task, particularly in environments with moving obstacles such as humans. We formulate this problem as a stochastic optimal control problem. Since solving the full problem is computationally demanding, we introduce a tractable approximation whose Bellman equation can be solved efficiently. The resulting value function is then incorporated as a terminal penalty in an online rollout framework. We construct a trade-off curve between safety and performance to identify an appropriate weighting between them, and compare the performance with other methods. Simulation results show that the proposed rollout approach can be tuned to reach the target in nearly the same expected time as receding horizon $A^\star$ while maintaining a larger expected minimum distance to the moving obstacle. The results also show that the proposed method outperforms the considered CBF-based methods when a larger obstacle clearance is desired, while achieving comparable performance otherwise.