The Critterbot Project is an initiative of the Reinforcement Learning and Artificial Intelligence (RLAI) lab at the University of Alberta. Research is primarily driven by reinforcement learning techniques; predictive state representation, dyna, temporal difference nets, options and option models, and others. Subjective representation of knowledge is a central theme, incoming sensor observations of the world are not filtered or normalized beyond the internal mechanisms of the sensors. We hope to gain experience in sensorimotor models in robotics, connecting low-level experiences and high-level knowledge, teaching and interaction, life-long learning, and real time reinforcement learning. Have a look at this video to know more.Crittergui can replay log files or start the Critterbot Simulator: an excellent platform for getting familiar with the Critterbot interface as well as stand alone research. The simulator is still in development but already has been used by several different groups.
Medias: demo of the Critterbot, all the videos, images.
The Critterbot is back to operational.
The Critterbot is currently down for an extended series of maintenance tasks. Michael Delp has solved the problem with the Wiimote, which is now back to operational.
Marc and Thomas had to reflash the AVRs. The robot should no longer be wandering around during the night.
The robot has a prototype docking station and beacon in the pen that allows it to charge itself once it's batteries are low, and this behavior has been successfully demonstrated. All originally planned sensors except for the bump sensors are operational and reporting full precision data. Stall protection and velocity limiting has been implemented on the motors.
The battery charger is functional and ready to use this week. There are some hardware issues with the new computer that among other things require it to be sent back for factory modifications, it will not be ready in the near future. July is slated to be the first long-run test of the platform, with everyone encouraged to work with it as much as possible. Many improvements in the firmware have made things much less buggy, though there is still some work to be done.
The robot is back together with some new hardware and the ability to charge its own batteries. Following the MSRL conference later this month the new on-board computer (Intel Atom based) will be installed, along with an updated Disco that should provide easier C++/Java communication.
The design for the on-board charger is being tested and a revised power board for the robot incorporating charging should be finished by mid-May. The first serious endurance tests for the platform put battery life at about 6-7 hours of constant use. Also in the works is a major revision to Disco as it approaches its first official release later this summer.
The Critterbot Mach I exists in hardware beta, many features are functional but a bit rough around the edges still. We have collected some data, available in the logs section, and will be adding more as testing continues. There is a 2.5 x 2.5m pen in the lab for the Critterbot to begin learning in a safe environment. The robot has recently demonstrated it's first learned behavior, a simple photovore agent, using tile coding. Many more experiments are slated for the near future