Vector Field-Guided Learning Predictive Control for Motion Planning of Mobile Robots with Uncertain Dynamics (2405.08283v3)

Abstract: In obstacle-dense scenarios, providing safe guidance for mobile robots is critical to improve the safe maneuvering capability. However, the guidance provided by standard guiding vector fields (GVFs) may limit the motion capability due to the improper curvature of the integral curve when traversing obstacles. On the other hand, robotic system dynamics are often time-varying, uncertain, and even unknown during the motion planning process. Therefore, many existing kinodynamic motion planning methods could not achieve satisfactory reliability in guaranteeing safety. To address these challenges, we propose a two-level Vector Field-guided Learning Predictive Control (VF-LPC) approach that improves safe maneuverability. The first level, the guiding level, generates safe desired trajectories using the designed kinodynamic GVF, enabling safe motion in obstacle-dense environments. The second level, the Integrated Motion Planning and Control (IMPC) level, first uses a deep Koopman operator to learn a nominal dynamics model offline and then updates the model uncertainties online using sparse Gaussian processes (GPs). The learned dynamics and a game-based safe barrier function are then incorporated into the LPC framework to generate near-optimal planning solutions. Extensive simulations and real-world experiments were conducted on quadrotor unmanned aerial vehicles and unmanned ground vehicles, demonstrating that VF-LPC enables robots to maneuver safely.

• The paper proposes a two-level VF-LPC framework that integrates kinodynamic vector fields with an adaptive motion planning and control level to generate safe trajectories.
• It combines a deep Koopman operator model with sparse Gaussian processes to address uncertainties and achieve near-optimal control.
• Experiments demonstrate VF-LPC’s superiority over MPC and RL methods in trajectory feasibility, route length, and computational efficiency.

Vector Field-Guided Learning Predictive Control for Mobile Robots with Unknown Dynamics

The paper "Vector Field-Guided Learning Predictive Control for Motion Planning of Mobile Robots with Unknown Dynamics" addresses significant challenges in the domain of mobile robotics, particularly in maintaining safety and optimal control in the presence of unknown dynamics and complex environments. The authors propose a two-level Vector Field-guided Learning Predictive Control (VF-LPC) framework that integrates motion planning and control, enhancing the maneuverability and safety of mobile robots through innovative methods.

The VF-LPC framework is structured into a guiding level and an integrated motion planning and control (IMPC) level. The guiding level employs a kinodynamic guiding vector field to generate safe trajectories within environments dense with obstacles. This vector field is adept at incorporating dynamic constraints, thereby eliminating common issues such as deadlock and ensuring that the curvature of generated paths remains feasible under the robot's dynamic constraints.

At the IMPC level, the paper introduces a novel approach combining offline learning with online adaptation. Initially, a deep Koopman operator is utilized to form a nominal dynamics model. To address model uncertainties and adapt to dynamic changes in real-world environments, the authors employ sparse Gaussian processes (GPs). This dual-level structure allows the IMPC to generate control sequences that are near-optimal and ensure safety through a game-based barrier function. The barrier function acts by dynamically switching between tracking and avoidance behaviors, thereby maintaining a safe distance from unexpected obstacles.

Quantitative evaluations reveal the efficacy of the VF-LPC framework. It outperforms existing model predictive control (MPC) and reinforcement learning (RL) methodologies across several metrics, including trajectory feasibility, completion time, route length, and computational efficiency. The paper describes simulated environments where VF-LPC was tested against established approaches like Hybrid A* and Timed Elastic Bands (TEB), showcasing its superior path length and solution time while maintaining safety.

Moreover, the authors validated VF-LPC through real-world experiments using a Hongqi E-HS3 vehicle. These experiments confirmed VF-LPC's effectiveness, illustrating its capability to adapt to system dynamics learning in practical scenarios.

Implications of this research extend both practically and theoretically. Practically, VF-LPC offers a scalable solution for deploying mobile robots in real-world settings where system dynamics are unknown and environmental conditions are unpredictable. Theoretically, it contributes to the field of model-based reinforcement learning by effectively integrating kinodynamic guidance with predictive control, presenting new directions for research in adaptive control systems.

Future research could extend VF-LPC's capabilities to handle higher speeds and more dynamic environments by integrating advanced prediction modules for obstacle behavior. Additionally, the framework's deployment on a broader range of robotic systems could further validate its adaptability and robustness, potentially inspiring advancements in autonomous vehicle technology and beyond.