Integrating Trajectory Optimization in Quality-Diversity for Kinodynamic Motion Planning

Kinodynamic planning is a critical component of robotics, enabling robots to generate dynamically feasible trajectories while adhering to geometric constraints. Existing methods for kinodynamic planning primarily fall into two categories: (a) stochastic planners, such as Rapidly-Exploring Random Trees (RRT) and Probabilistic Roadmaps (PRM), which excel in state-space exploration due to their directed randomness, and (b) optimization-based approaches, which are well-suited for structured environments, producing smooth and optimal solutions. In this work, we present Hybrid Trajectory Exploration for Kinodynamic Planning (HyTraX), a hybrid framework that integrates the strengths of these two paradigms. Specifically, HyTraX enhances the effectiveness of the Go-Explore algorithm by incorporating trajectory optimization with random target selection as its core exploration strategy. We evaluate HyTraX on two challenging kinodynamic motion planning tasks: a car-like agent and a planar quadrotor navigating through maze environments. Our results demonstrate significant performance improvements with minimal task-specific tuning, highlighting the robustness and versatility of the proposed approach.