Iterative MPPI

My current work is based on the original MPPI(Model Predictive Path Integral Control) method proposed by ACDS-Lab in Georgia Tech.

The original MPPI algorithm is a sampling-based type of MPC algorithms, and it can be derived using different methodologies that include, information theoretic dualities between free-energy and relative-entropy, stochastic search approaches and variational optimization methodologies. I took information-theory-based MPPI for my research.

My original thought about this research is to do something like MPPI game with multi-agent interaction, and in order to acheive this, directed by Dr. Jun Zeng, I’m doing the following prestage work: Inspired by the iterative framework of our former work, I’d like to find a policy to perform MPPI iteratively in scenarios with obstacles and automatically work out the tendency of avoiding the obstacle. It should be a combination of scenario generator, planner and controller. The policy is under testing stage now.