Autonomous Navigation of <i>AGV</i>s in Unknown Cluttered Environments: <i>Log-MPPI</i> Control Strategy

Sampling-based model predictive control ( MPC ) optimization methods, such as Model Predictive Path Integral xmlns:xlink="http://www.w3.org/1999/xlink">MPPI ), have recently shown promising results in various robotic tasks. However, it might produce an infeasible trajectory when the distributions of all sampled trajectories are concentrated within high-cost even regions. In this study, we propose a new method called xmlns:xlink="http://www.w3.org/1999/xlink">log-MPPI equipped with more effective sampling distribution policy which significantly improves feasibility terms satisfying system constraints. The key point is to draw samples from normal log-normal xmlns:xlink="http://www.w3.org/1999/xlink">NLN mixture distribution, rather than Gaussian distribution. Furthermore, work presents for collision-free navigation unknown cluttered environments by incorporating xmlns:xlink="http://www.w3.org/1999/xlink">2D occupancy grid map into problem sampling-based algorithm. We first validate efficiency and robustness our proposed strategy through extensive simulations autonomous different types well cartpole swing-up task. further demonstrate, real-world experiments, applicability performing grid-based environment, showing its superiority be utilized local costmap without adding additional complexity problem. A video demonstrating simulation available at https://youtu.be/_uGWQEFJSN0 .