Path Morphing for a Single Robot Performing a Sensing Task or a Persistent Task

From DRLWiki

Persistent robot tasks such as monitoring and cleaning are concerned with controlling mobile robots to act in a changing environment in a way that guarantees that the uncertainty in the system (due to change and to the actions of the robot) remains bounded for all time. Prior work in persistent robot tasks considered a robot traveling along a pre-defined path, and having exact knowledge of the environment in order to calculate the stabilizing speed profile for the persistent task. In real-life situations a path might not be known and the environment might also be unknown. This work is intended to alleviate these assumptions.

In this work an adaptive control law is generated to drive a collection of waypoints of a single robot path to a near-optimal configuration for sensing. The final location of the waypoints must generate a “smooth” closed path for a robot to follow in order to sense the points of interest in the environment to perform, for example, a persistent task. A Lyapunov-like stability proof using a generalized invariant set theorem was used to show the convergence of the system to an equilibrium point under the proposed adaptive control law. Although the proposed approach is useful for any sensing task, an extension to the controller was designed, which drives the system to a state that is beneficial to performing stable persistent tasks. Simulation results support the proposed approach. Experiments using quadrotors will be performed soon.

Approach

Building upon previous work by Mac Schwager, et al. on Voronoi-based coverage, an adaptive controller was designed to drive a collection of waypoints defining the robot's path into an informative path. By informative path we mean a path that allows the robot to sense information-rich locations in the environment. These waypoints have a relationship with the previous and next waypoints, in which they want to stay close to each other, creating a closed path. The controller drives the waypoints to an equilibrium point where it balances between going to the Voronoi centroid and staying close to its neighbor waypoints.

In order to treat unknown environments, an adaptation law was generated that drives the robot's estimate of the environment to the real environment at all locations it visits, making the robot's estimate be a good representation of the real environment in the visited locations.

In order to use our informative paths for persistent sensing, we use a speed controller from previous work to calculate the speed along each point in our informative path in order to persistently sense the dynamic environment.

The initial waypoint configurations of a simulation. The waypoints are the blue circles, the path can be seen by the black line connecting all waypoints. The blue lines are shown to view the initial Voronoi cells where the waypoints are the generator points. A sample of the points of interest in the environment are marked by red dots. The initial configuration is designed to cover most of the environment so that the robot's estimate converges to the real environment for all of the environment.

The final waypoint configurations of a simulation. The waypoints are the blue circles, the path can be seen by the black line connecting all waypoints. The blue lines are shown to view the final Voronoi cells where the waypoints are the generator points. A sample of the points of interest in the environment are marked by red dots. The achieved final configuration covers all points of interest in the environment, and enables the robot to perform a successful persistent task.

This approach was then extended to the multi-robot case, where multiple robots have to work together to persistently sense the unknown dynamic environment. A similar controller and adaptation law were generated to solve this multi-robot case.

Error creating thumbnail:

The initial configuration for a 2-robot experiment. Green dots represent points of interest and the size of the dots represents the accumulation function related to the persistent task. The goal of the robots is to maintain the size of these dots small for all time. Both robots start off with an arbitrary initial path (red for one robot and blue for the other).

Error creating thumbnail:

Configuration after 50 iterations of the controller for a 2-robot experiment.

Error creating thumbnail:

Final configuration for a 2-robot experiment. The final paths, combined with the speed along the paths generate informative trajectories for both robots that allow them to persistently sense the unknown dynamic environment.

Publication

Single-robot case D. E. Soltero, M. Swchager, and D. Rus, “Generating informative paths for persistent sensing in unknown environments,” in Intelligent Robots and Systems (IROS), 2012 IEEE/RSJ International Conference on, 2012.

Multi-robot case Submitted to ICRA 2013.

People

Daniel E. Soltero (MS/Ph.D. student, MIT)

Mac Schwager (Assistant professor, Boston University)

Daniela Rus (Professor, MIT)