OAA Release 2 coming soon

Release 2 will contain many of new features, including:

OAA-Enabled Robots Wins AAAI Robotics Contest!

Report on the SRI Pioneer Robot Team at AAAI96 Robotics Competition and Exhibition

At the SRI Artificial Intelligence Lab, we have a long history of building autonomous robots, from the original Shakey (remember the STRIPS planner?) thru Flakey and more recently, the Erratic/Pioneer class of small robots. Our current research focuses on realtime vision for robots, and multirobot planning using an agent-oriented architecture.

For the AAAI contest, we wanted to showcase our research, especially the ability to control multiple robots using a distributed set of software agents on the Internet. The agent technology, called the Open Agent Architecture^TM (OAA^TM), was developed at SRI as a way of accessing information available in computers at many different locations.

In the "Call a Meeting" event, a robot starts from the Director's office, determine which of two conference rooms is empty, notifies two professors where and when the meeting would be held, and then returns to tell the Director. Points are awarded for accomplishing the different parts of the task, for communicating effectively about its goals, and for finishing the task quickly. Our strategy was simple: use as many robots as we could to cut down on the time to find the rooms and notify the professors. We decided that three robots was an optimal choice: enough to search for the rooms efficiently, but not too many to get in each other's way or strain our resources. We would have two robots searching for the rooms and professors, and one remaining behind in the Director's office and tell her when the meeting would be.

The two search robots are Pioneer-class robots, portable robots manufactured by Real World Interface and running SRI's Saphira control software. Saphira navigates the robots around an office environment, keeping track of where they are using perceptual cues from the robot sensors. Each Pioneer robot has seven sonar sensors, a Fast Track vision system from Newton Labs, and a portable computer on top with a Netwave radio ethernet for communication to a base station. We decided to use the vision system to find people in the conference rooms: if the judges would wear red shorts, the vision system would easily pick them out.

The robot in the Director's office didn't have to move, just relay information to the Director. We used a Koala robot, a small 6-wheeled vehicle under development at the Swiss Federal Institute.

Each robot, by virtue of the radio ethernet, is a node on the internet and an agent in the OAA. Other agents, residing on the base station, included a database agent for holding information relayed back by the robots, a mapping agent for determining and displaying where the robots are, a strategy agent for coordination, and interface agents (speech, pen gestures) for giving the robots commands. If we had a connection to the outside world, we could have run the robots from anywhere in the world!

One of the interesting aspects of the agent architecture is that the robots are capable of a good deal of autonomy. For example, the strategy agent might tell them where to go, and even give them a path. The robots are responsible for navigation, avoiding obstacles, and finding the correct goal position. If they fail, they contact the strategy agent about the problem and wait for instructions. The connection between the robots and the rest of the agents can be very low bandwidth.

This was our strategy; now we had to execute it. We arrived at the contest late, during the preliminary rounds, and hastily set up our base station and put in the map of the office environment. All did not go smoothly --- several unanticipated coordination problems between the mapping agent and the robots caused us some frustrating moments. We hadn't realized the planner would happily plan paths through rooms with two doors, something the robots didn't like because they get lost easily in rooms and do much better in the corridors. And one of our robots was injured in a fall as we were packing up at SRI; fortunately RWI was there, and loaned us another Pioneer. And so, late in the evening of the second day, we had a perfect preliminary round run: the two search robots each went to a different conference room, found the empty one, then went to different professors' offices and told them where the meeting was.

The next day, the finals, was almost an anticlimax, except for the large audience and the presence of Alan Alda and the camera crew from Scientific American Frontiers. We started the Pioneers out from the Directors office, and they happily went on their way. There was too much going on to keep track of both of them --- they were both announcing what they were doing, the camera folks were dancing around, and we had to tell the audience what was happening. Before we knew it, both robots were in the professors' offices, announcing the meeting: only 4 minutes and 30 seconds to completion! From watching the other teams, we knew the nearest time would be close to 10 minutes. Parallelism does work sometimes...

Our only disappointment was not being able to show off SRI's Small Vision Module, which is a credit-card size system consisting of two small cameras, digitizers, and a processor. It performs realtime stereo for obstacle avoidance and people detection, but was not quite ready to be fitted to the robots. Next year.

Kurt Konolige
August 22, 1996