Application of time dependent probabilistic collision state checkers in highly dynamic environments

Area of Science:

• Robotics and Artificial Intelligence
• Autonomous Systems Navigation

Background:

• Autonomous vehicle trajectory planning must avoid collisions with static and dynamic obstacles.
• Treating collisions as binary events is insufficient in unpredictable, dynamic environments.
• Predicting the future positions of moving obstacles is crucial for safe navigation.

Purpose of the Study:

• To develop a time-dependent probabilistic collision state checker for autonomous vehicles.
• To enable robots to trace statistically safe trajectories in crowded, dynamic environments.
• To minimize collision probability during autonomous vehicle path planning.

Main Methods:

• Utilized a sequential Bayesian model for approximate predictions of obstacle movement patterns.
• Defined a time-dependent variation of the Dijkstra algorithm for trajectory computation.
• Integrated probabilistic collision checking into the path planning process.

Main Results:

• Demonstrated the ability to compute statistically safe trajectories.
• Achieved minimum collision probability routes in simulated crowded environments.
• Validated the efficiency of the proposed methods through experimentation.

Conclusions:

• The developed system enhances the safety of autonomous vehicles in dynamic settings.
• Probabilistic collision checking offers a more robust approach than binary methods.
• The integration of predictive modeling and modified Dijkstra algorithm proves effective for safe robot navigation.