Anticipative Kinodynamic Planning: Robot Navigation in Urban and Dynamic Environments

Gonzalo Ferrer and Alberto Sanfeliu

Abstract- This paper presents the Anticipative Kinodynamic Planning (AKP) approach for robot navigation in crowded urban environments, while satisfying both dynamic and nonholonomic constraints. To this end, we propose to integrate seamlessly a human motion prediction algorithm into the planning algorithm. The AKP generates a set of paths in a kinodynamic RRT fashion and choses, among a set of candidates, the best path. We go one step further and formulate the problem from a multi-objective perspective, where we seek to optimize different and independent measures, such as, the robot cost, pedestrians trajectories costs, etc... Additionally, a multi-objective cost-to-go function is proposed to measure the distance between states, which results in an improved construction of the planner tree. We have dramatically speed up the calculations by using a steering heuristic and at the same time, we have introduced a randomness factor to the steer method, to obtain more distinct paths, which is of great importance for the optimization process. Simulations and real experiments have been carried out to demonstrate the success of the Anticipative Kinodynamic Planner.


Videos In the following videos is depicted the performance of the Anticipative Kinodynamic Planning (AKP), our robot navigation approach:

Environment: We show the main components of our planner GUI.

Simulations: AKP simulation environment consisting of a different number of pedestrians and obstacles where the algorithm samples parameters in order to learn their value after thousands of simulations.

Experiments: The AKP navigation is compared to other methods in a simple and controlled environment.

Testing: The AKP is tested in uncontrolled environments, that is, no instructions were given to the pedestrians in the area while the robot was navigating

Derived publications:

G. Ferrer and A. Sanfeliu. "Proactive Kinodynamic Planning using the Extended Social Force Model and Human Motion Prediction in Urban Environments". In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 1739-1735. Chicago, USA, 2014. [info] [pdf] [video]

You can check all our code soon.

_