MORE | Spring 2024

Unified Frenet Frame Motion Planning Based Framework for Social Navigation

Data icon, disabled. Four grey bars arranged like a vertical bar chart.

Understanding AV-human interaction is crucial yet challenging, often leading to accidents. Game theory optimizes this but is computationally inefficient. Monte Carlo Tree Search (MCTS) improves efficiency but exhaustive exploration remains an issue. Pruning, as seen in AlphaGo, helps. The researchers’ proposed framework combines MCTS with predictive motion from frenet frame-based planning, focusing on specific vehicle states. Validation with real-time data, from INTERACTION dataset, enhances reliability. The researchers transform the 2D dataset into frenet frame predictions, deploy actions in a Stackleberg game using Sarsa (lambda) and manipulate the way reward functions backpropagate.

Student researcher

Karthick Subramanian

Robotics and autonomous systems

Hometown: Chennai, Tamil Nadu, India

Graduation date: Spring 2024