Advanced Visual Predictive Control Scheme for the Navigation Problem

This work proposes a Visual Predictive Control (VPC) scheme adapted to the autonomous navigation problem among static obstacles. To do so, it is necessary to cope with several issues which by now limit the use of VPC in this context. Among them, we focus on the following ones: the need for precise prediction models to improve the task realization; the need for a long prediction horizon which is required to perform long range displacements and guarantee stability, but also results in a high computational burden and a more difficult implementation ; the possible optimization problem evolution at every iteration due to unexpected events (e.g., detection of new obstacles), which leads to non convex problems and therefore makes difficult its resolution. The proposed VPC allows to tackle the above mentioned challenges. Based on a more accurate prediction model relying on an exact integration method, it integrates constraints to deal with actuator saturation, obstacle avoidance along the trajectory and stability. To deal with the two last mentioned challenges, the classical VPC scheme has been extended with two methods: the first one allowing to relax some constraints on the control inputs to reduce the computational burden; the second one for adequately refining the optimized trajectory to avoid local minima when the optimization problem evolves during the navigation. The proposed approach has been evaluated and compared to other VPC configurations. The obtained results show than it runs 60 times faster than classical configurations for similar performances.

Automated summary

This work proposes a Visual Predictive Control (VPC) scheme adapted to the navigation problem among static obstacles, addressing challenges such as accurate prediction models, long prediction horizon, and optimization problem evolution. The proposed approach improves prediction accuracy, integrates constraints, refines optimized trajectory, and achieves significantly faster performance compared to classical configurations.