Planning in mobile robots
A computer system for planning mobile robot trajectories, the computer system comprising: an input configured to receive a set of scenario description parameters describing a scenario and a desired goal for the mobile robot therein; a runtime optimizer configured to compute a final mobile robot trajectory that substantially optimizes a cost function for the scenario, subject to a set of hard constraints that the final mobile robot trajectory is guaranteed to satisfy; and a trained function approximator configured to compute, from the set of scenario description parameters, initialization data defining an initial mobile robot trajectory. The computer system is configured to initialize the runtime optimizer with the initialization data, in order to guide the optimizer from the initial mobile robot trajectory to the final mobile robot trajectory that satisfies the hard constraints, the function approximator having been trained on example sets of scenario description parameters and ground truth initialization data for the example sets of scenario description parameters.
1 . A computer system for planning mobile robot trajectories, the computer system comprising one or more hardware processors configured to:
receive a set of scenario description parameters describing a scenario and a desired goal for a mobile robot in the scenario;
implement a runtime optimizer to compute a final mobile robot trajectory for the mobile robot that substantially optimizes a cost function for the desired goal in the scenario, subject to a set of hard constraints for the scenario that the final mobile robot trajectory is guaranteed to satisfy;
implement a function approximator to compute, from the set of scenario description parameters, initialization data defining an initial mobile robot trajectory for the mobile robot;
initialize the runtime optimizer with the initialization data;
guide the runtime optimizer from the initial mobile robot trajectory to the final mobile robot trajectory that satisfies the set of hard constraints, the function approximator having been trained on example sets of scenario description parameters and ground truth initialization data for the example sets of scenario description parameters;
generate a series of control signals for controlling the mobile robot to follow the final mobile robot trajectory; and
control motion of the mobile robot to follow the final mobile robot trajectory based on the series of control signals.
2 . The computer system of claim 1 , wherein the function approximator has been trained to approximate a reference planner, the ground truth initialization data having been generated by applying the reference planner to the example sets of scenario description parameters.
3 . The computer system of claim 1 , wherein the runtime optimizer is configured to determine a series of control actions, and compute the final mobile robot trajectory by applying a robot dynamics model to the series of control actions.
4 . The computer system of claim 3 , wherein the initialization data comprises an initial sequence of control actions defining the initial mobile robot trajectory.
5 . The computer system of claim 1 , wherein the set of hard constraints comprise one or more collision avoidance constraints for one or more static or moving obstacles in the scenario, and location(s) of the one or more static or moving obstacles are encoded in the set of scenario description parameters for use by the function approximator.
6 . The computer system of claim 1 , wherein the set of hard constraints comprise one or more permitted area constraints for keeping the mobile robot within a permitted area, and the permitted area is encoded in the set of scenario description parameters for use by the function approximator.
7 . The computer system of claim 1 , wherein the desired goal is defined relative to a reference path, and the cost function encourages achievement of the desired goal by penalizing at least one of lateral deviation from the reference path, and longitudinal deviation from a reference location on the reference path.
8 . The computer system of claim 7 , wherein the initial and final trajectories are represented in a frame of reference defined by the reference path.
9 . The computer system of claim 1 , wherein the function approximator has a convolutional neural network (CNN) architecture.
10 . The computer system of claim 9 , wherein the set of hard constraints comprise:
collision avoidance constraints for one or more static or moving obstacles in the scenario, wherein location(s) of the one or more static or moving obstacles are encoded in the set of scenario description parameters for use by the function approximator; or
one or more permitted area constraints for keeping the mobile robot within a permitted area, wherein the permitted area is encoded in the set of scenario description parameters for use by the function approximator;
wherein the computer system is configured to transform the set of scenario description parameters into an input tensor comprising one or more images visualizing the permitted area or the location(s) of the obstacles.
11 . The computer system of claim 10 , wherein the set of hard constraints comprise collision avoidance constraints for one or more static or moving obstacles in the scenario, wherein the input tensor comprises multiple images visualizing predicted locations of the obstacles at different time instants.
12 . The computer system of claim 10 , wherein the desired goal is defined relative to a reference path, and the cost function encourages achievement of the desired goal by penalizing at least one of lateral deviation from the reference path, and longitudinal deviation from a reference location on the reference path, and wherein the one or more images encode the reference path.
13 . The computer system of claim 12 , wherein the initial and final trajectories are represented in a frame of reference defined by the reference path, wherein the one or more images encode the reference path by visualizing the permitted area or the location(s) of the obstacles in the frame of reference defined by the reference path.
14 . The computer system of claim 1 , wherein the function approximator encodes the initial mobile robot trajectory as a set of smooth function parameters.
15 . The computer system of claim 1 , wherein the runtime optimizer is a non-linear optimizer.
16 . A computer program embodied on non-transitory computer-readable media for programming one or more computers to implement steps of:
receiving a set of scenario description parameters describing a scenario and a desired goal for a mobile robot in the scenario;
computing, by applying a function approximator to the set of scenario description parameters, initialization data defining an initial mobile robot trajectory for the mobile robot; and
computing, by applying a runtime optimizer, a final mobile robot trajectory for the mobile robot that substantially optimizes a cost function for the desired goal in the scenario, subject to a set of hard constraints for the scenario that the final mobile robot trajectory is guaranteed to satisfy,
initializing the runtime optimizer with the initialization data;
guiding the runtime optimizer from the initial mobile robot trajectory to the final mobile robot trajectory that satisfies the set of hard constraints, the function approximator having been trained on example sets of scenario description parameters and ground truth initialization data for the example sets of scenario description parameters;
generating a series of control signals for controlling the mobile robot to follow the final mobile robot trajectory; and
controlling motion of the mobile robot to follow the final mobile robot trajectory based on the series of control signals.