Flagship research
FFStreams / FFStreams++
From FFStreams task-and-motion planning to FFStreams++ prediction-aware maneuver planning, developed during my Ph.D.
Ph.D. research · MIPT · Algorithm developer · First author
FFStreams++ demonstrations
Passing an intersection
Unprotected left turn
The problem
Autonomous vehicles must choose safe maneuvers while accounting for vehicle constraints and the possible future movements of surrounding traffic.
My contribution
- I developed FFStreams during my Ph.D., combining heuristic search with Streams for task-and-motion planning.
- I then developed FFStreams++, extending the approach with QCNet multi-agent trajectory prediction for prediction-aware maneuver planning.
- I evaluated maneuver planning in driving scenarios, including intersection passing and unprotected left turns.
Technical approach
I first developed FFStreams to connect high-level maneuver decisions with low-level motion planning through heuristic search and Streams. I then developed FFStreams++, incorporating predicted trajectories of surrounding vehicles into maneuver planning. The demonstrations and experiment plots below show the FFStreams++ stage of this research.
Results & validation
Evaluated on CommonRoad driving scenarios
These two FFStreams++ experiments show intersection passing and an unprotected left turn. Each result figure pairs the maneuver sequence with velocity, acceleration, and jerk over time, with YIELD and FOLLOW decision annotations. They illustrate behavior in individual scenarios; the linked papers provide the broader experimental evaluation.
Intersection passing — experiment results
Unprotected left turn — experiment results
