The LASR lab offers a wide variety of capabilities in Ground-Based Robotics. Aside from the large HOMER robot used for proximity operations, the lab also maintains a small fleet of low-cost iRobot Creates, which are used as a simple platform to simulate and validate control algorithms with hardware in the loop.
Pursuit-Evasion games represent a specific class of optimal control problems in which both the pursuer and evader attempt to play optimally by evaluating some cost function defining the game and their intentions. In a zero-sum game, the pursuer tries to minimize the same cost function that the evader is trying to maximize. A saddle point (if it exists) in a zero-sum game defines the best outcome any player can hope to achieve when the opponent is also playing optimally. It's possible to realize the theoretical outcomes to these optimal control problems in hardware by tailoring the solution strategy to exploit the differentially flat characteristics of a planar robot.
However, any realistic scenario between uncooperative players has inherent strategy uncertainty. When one or more players plays suboptimally or passively, the expected outcome changes drastically. To account for this strategic uncertainty, methods such as behavior learning and receding horizon control (RHC) or model predictive control (MPC) can be used to augment a player's strategy.
The one pursuer, one evader game has been successfully implemented in the laboratory with each player assigned an iRobot Create outfitted with an onboard netbook computer or Raspberry Pi. The Vicon motion capture system provides the centralized control computer with state information of each player. An optimal solution has been demonstrated where offline controller computation based on the defined game is performed, then commands are sent wirelessly to each player. Suboptimal solutions have been implemented and run onboard for online real-time solution. Current efforts are being focused on implementing the two pursuer, one evader game. In this particular scenario, the pursuers have the option of working cooperatively or independently. Additional work has been done toward implementing behavior learning and RHC methods to account for strategic uncertainty.
The LASR Lab boasts 10 custom-equipped quadrotors to study cooperative and adaptive control laws for unmanned aerial vehicles. Under the auspices of the Texas A&M Space Engineering Institute, with sponsorship from Boeing, a team of undergraduates designed a custom wireless control circuit board, complete with acceleration and rotation sensors. The team also outfitted the aircraft with a custom superstructure to host Vicon reflectors and PhaseSpace beacons, thus enabling high-frequency tracking of the vehicle through our sensing systems.
Recent efforts include the incorporation of a prototype autonomous framework and the AR Drone vehicle into landing site determination experiments that require an aerial component. Currently, new methods for autonomous control of the quadrotors are being developed. Future goals include use in pursuit-evasion testing and vision sensor integration.
Humberto Ramos, Tim Woodbury, and Austin Probe