High-Speed Navigation Using Onboard Vision
Recent advances in electro-optical cameras afford a unique opportunity for lightweight high-frequency sensing of the environment. However, as the speed of robot systems increases, the uncertainty associated with these sensor systems is more likely to negatively impact system performance. Noisy pose estimates from visual odometry, motion blur, and lighting conditions can all contribute significantly to uncertainty. Furthermore, limited field of view (FOV) can lead to overly conservative navigation speeds.
To overcome these challenges, ISC researchers explored integrating their nonlinear model predictive control (NMPC) approaches with computationally lightweight mapping frameworks. In addition, we have also explored machine-learning algorithms such as generative adversarial networks to predict and reason about the environment beyond the sensor’s FOV for more robust and high-speed navigation.
Fast Motion Planning Using PAC-NMPC
NMPC is typically restricted to short, finite horizons to limit the computational burden of online optimization. This makes a global planner necessary to avoid local minima when using NMPC for navigation in complex environments. For this reason, the performance of NMPC approaches is often limited by that of the global planner. While control policies trained with reinforcement learning (RL) can theoretically learn to avoid such local minima, they are usually unable to guarantee enforcement of general state constraints. In this paper, we augment a sampling-based stochastic NMPC (SNMPC) approach with an RL-trained perception-informed value function. This allows the system to avoid observable local minima in the environment by reasoning about perception information beyond the finite planning horizon. By using Probably Approximately Correct NMPC (PAC-NMPC) as our base controller, we are also able to generate statistical guarantees of performance and safety. The approach was demonstrated in simulation and on hardware using a 1/10th-scale rally car with lidar.
- Adam Polevoy, Marin Kobilarov, Joseph Moore, “Probably Approximately Correct Nonlinear Model Predictive Control (PAC-NMPC),” IEEE Robotics and Automation Letters (RA-L), Vol. 8, Issue 11 (Nov 2023).