Unifying Decision Making and Trajectory Planning in Automated Driving through Time-Varying Potential Fields
A new AI framework merges decision-making and trajectory planning, handling dynamic obstacles with bounded uncertainty.
A team of researchers has published a paper proposing a novel, unified framework for autonomous vehicle navigation. The core innovation is the use of Time-Varying Artificial Potential Fields (TVAPFs), which model the predicted motion of dynamic obstacles—like other cars and pedestrians—by accounting for their bounded uncertainty over time. This model integrates data from onboard perception systems and, when available, Vehicle-to-Everything (V2X) communication. By explicitly forecasting how obstacles might move, the system can plan safer, more anticipatory paths.
The TVAPF model is embedded into a finite-horizon optimal control problem. This formulation allows the system to simultaneously solve two critical tasks: high-level decision-making (e.g., deciding to change lanes or brake) and low-level trajectory planning (plotting the exact path and speed). The result is a single, optimized output that selects a driving maneuver and computes a specific, feasible, and collision-free trajectory. The researchers demonstrated the framework's effectiveness and computational efficiency through simulation tests in complex, multi-actor scenarios based on real road layouts, highlighting advantages over decoupled planning approaches.
- Proposes Time-Varying Artificial Potential Fields (TVAPF) to model dynamic obstacle motion with bounded uncertainty.
- Unifies high-level decision-making and low-level trajectory planning into a single optimal control problem.
- Validated in simulation with real road topology, showing real-time performance in multi-actor scenarios.
Why It Matters
This unified approach could lead to safer, more efficient, and more predictable autonomous driving systems by closing the gap between decision logic and motion control.