Spline-Based Multi-Level Planning for Autonomous Vehicles Rahul Kala The paper was extended and published as: R. Kala, K. Warwick (2013) Multi-Level Planning for Semi-Autonomous Vehicles in Traffic Scenarios based on Separation Maximization, Journal of Intelligent and Robotic Systems, 2013,DOI:10.1007/s10846-013-9817-7 01 December 2018
Autonomous Vehicles Safety Efficient Driving Comfort Jam Avoidance Coordination Comfort
Conventional Model Planning
Why not speed lanes? Coordination Highly Diverse Speeds Highly Diverse Sizes
Why not speed lanes? Single lanes And if highly crowded
Why not speed lanes? “Our model assumes that vehicles travel only along lanes or on certain lane-change path. In California, the practice of “lane-splitting” is legal — motorcycles are free to travel in between cars in adjacent lanes. This occurs in the I-80 dataset, and presents a challenge for our method, which must try to find a path around such obstacles and force each vehicle to precisely follow a single lane.” –Sewall et al. (2011) J. Sewall, J. van den Berg, M. C. Lin, D. Manocha, D, “Virtualized Traffic: Reconstructing Traffic Flows from Discrete Spatiotemporal Data”, IEEE Transaction on Visualization Computer Graphics, 17(1), 26-37 (2011).
Why not conventional Path Planning? Pre-known/same time of emergence Wide spaces around High mobility/Low Speeds
From Literature Source: R. Kala, et al., Robotic path planning in static environment using hierarchical multi-neuron heuristic search and probability based fitness, Neurocomputing (2011), doi:10.1016/j.neucom.2011.03.006
From Literature Map Level 1 Level 2 Source: R. Kala, et al., Fusion of probabilistic A* algorithm and fuzzy inference system for robotic path planning, Artificial Intelligence Review, Vol. 33, No. 4, pp 275-306
Multi-Level Planning
Results
Results – Single Vehicle Scenarios
Results – 2 Vehicle Scenarios
Results – Multi- Vehicle Scenarios
Results - Overtaking
Results – Path Following
Solution Vehicle to be planned Road/Crossing Map Road Selection Path Pathway Selection Pathway Distribution Trajectory Generation Vehicle to be planned Road/Crossing Map Path Pathway Distributed Pathway Trajectory Replan All Vehicle Pathways All Vehicle Trajectories Controller
Road Selection
Separation Maximization Pathways Hypothesis from: J. R. Alvarez-Sanchez, F. de la Paz Lopez, J. M. C. Troncoso, D. de Santos Sierra, “Reactive navigation in real environments using partial center of area method”, Robotic and Autonomous Systems, 58(12), 1231-1237 (2010).
Pathway Selection
Pathway Selection Dijkstra’s algorithm cost ds(Pajk(m2)) = ds(Pajk(m1)) + || end(Pajk(m2)) – end(Pajk(m1)) || min_width(Pajk(m2)) = min(width(Pajk(m2)), min_width(Pajk(m1)),wmax) cost(Pajk(m2)) = ds(Pajk(m2)) + α min_width(Pajk(m2))
Coordination and Re-planning Ri is said to have a higher priority compared to Rr if Ri and Rr are driving in same direction of road and Ri lies ahead of Rr. Or Ri and Rr are driving in opposite directions of road and point of collision lies in left side of complete road.
Pathway Distribution Separation Pathways
Pathway Distribution Pathway Distribution Vehicle 2 (Speed=5) Overtake Vehicle 3 (Speed=15) Pre-preparation
Pathway Distribution Prepare yourself early for distribution change - Pre-preparation Late change of distribution - Post-preparation
Coordination and Re-planning Ri has a higher priority if It lies ahead of Rr with Ri and Rr going in same direction Or Rr and Ri have different directions.
Trajectory Generation
Trajectory Generation
Trajectory Generation Vehicle 2 (Speed=5) Vehicle 1 (Speed=5) Vehicle 3 (Speed=15)
Thank You