Randomised parameterisation motion planning for autonomous cars

Motion planning is the development of a set of continuous, executable trajectories, to guide a robot (a passenger vehicle in this case) from a current state towards a desired, goal state. Traditional planning algorithms are limited to simplified motion modes and environments. Sampling based planners are a recent development in robotic research. They rely on randomized exploration of the robot’s environment. Sampling based planners were successfully used for robotic planning amongst many other applications. Current planners are not suitable for autonomous passenger vehicle planning. The primary objective of the presented research is to develop novel methods and algorithms for sampling-based planning approaches. As such, those methods are expected to overcome compromises of existing planners. This is achieved by integrating a spline formulation of the vehicle’s motion within an efficient and consistent randomized planner. Splines are capable of synthesizing complex shapes whilst maintaining various degrees of parametric continuities. They are examined to devise a formulation for generating kinematically feasible paths and maintaining continuous trajectories. The continuity of the developed paths was assessed by comparing their parametric continuity with existing discontinuous models. Data sets with varied continuity classes were generated for standard manoeuvres. The feasibility of the resulting paths was verified using a bicycle model for front wheel steered vehicles. Implementation of different tracking controllers was used to evaluate improvements in path tracking performance and reported reductions in tracking error and control effort. Similarly, an evaluation of the steering disturbances under ideal and stochastic actuation conditions was conducted and revealed reductions in lateral acceleration. Results were validated using field experiments on a specifically developed robotic platform. Reduced disturbances and improved tracking performances are expected to improve vehicles’ stability, passenger comfort and reduce mechanical failure rate and tire wear. We validated the advantage of parametric path continuity on path tracking performance and passenger comfort. Based on the established spline parameterisation, randomised planning algorithms were developed that used spline paths for configuration space exploration. The resulting paths were validated for continuity and kinematic feasibility. Benchmarking against state of the art planners was conducted. Standard experiments were performed in maze environments, field environments and structured on-road environments. Nonparametric statistical analysis was used to evaluate planning time results. Proposed algorithms achieved significant improvements in planning time performance, compared to state of the art randomized kinodynamic spline based planners. In conclusion, motion safety and passenger comfort were identified as the primary challenges of autonomous passenger vehicle motion planning. A spline parameterisation based model of the vehicle’s motion is developed with high order parametric continuity. Application of this model improves path tracking performances and reduces resulting disturbances. Finally, the integration of the proposed spline parameterization, within the developed randomized search algorithm, achieves statistically significant improvements of autonomous passenger cars path planning performances. The proposed approach is capable of solving motion planning problems for autonomous cars in order of milliseconds for a variety of maze, field and on-road benchmarks. The outcomes contribute to enabling the development of provably safe and reliable self driving vehicles, in order to enhance intelligent transportation systems