Autonomous Driving Speed Planning via ST Graph Gap Selection

The patent segments the continuous speed planning problem into discrete speed segments corresponding to different trajectory profiles. By dividing the speed domain into multiple profiles (e.g., aggressive, moderate, conservative), the non-convex optimization is transformed into a sequence of simpler convex QP subproblems, each handling a specific speed profile range, thereby reducing computational complexity while maintaining synchronization accuracy.

The patent introduces an intermediary layer between path planning and speed planning by using the ST graph as a bridge. The ST graph discretizes the continuous space-time domain into viable cells, serving as an intermediary representation that captures space-time constraints. This intermediary structure allows the optimization algorithm to work with discrete profiles rather than continuous functions, simplifying the computational burden.

The patent extracts the speed planning dimension from the full state-space optimization problem. Instead of optimizing over all states and control inputs simultaneously as DP would require, the invention separates speed optimization from position optimization. By taking out the speed profile as an independent variable and optimizing it separately through QP, the method avoids the exponential dimensionality curse of DP while maintaining optimization precision.

The patent transforms the optimization problem from the state-space domain to the speed-profile domain. By introducing a new dimension of speed profiles and formulating the problem in terms of selecting and optimizing these profiles rather than optimizing individual state transitions, the method achieves precise optimization without being constrained by the dimensional limitations of traditional DP approaches.

The patent applies dynamics by treating speed as a dynamic variable that evolves through discrete profiles rather than a static parameter. The system dynamically selects and transitions between different speed profiles based on current state and constraints. This dynamic approach allows seamless handling of zero-speed scenarios by switching to profiles that accommodate stopping, avoiding the infinite time evaluation problem that arises in static space-domain planning.

The patent changes the parameterization of the planning problem from pure space-domain coordinates to a hybrid representation that includes speed profile parameters. By parameterizing trajectories in terms of speed profiles (which encode both spatial and temporal information), the method naturally handles position-based speed limits while avoiding singularities. The speed profile parameters provide a robust framework that remains well-defined even when velocity approaches zero, unlike time-based parameterizations.