The Maslab arena.MIT hosts an awesome robotics competition every January called Maslab. It's just for undergrads ,and the participants actually earn school credit. This is not a typical robotics contest. In the spirit of MIT, conestants are given less than 30 days to design and build a vision-based autonomous robot. The Maslab team gave mini-lectures and had regular checkpoints to ensure that everyone had something to show at the final contest.
The robots were powered by small, Linux based computers and used the OrcBoard to communicate with actuators and sensors. The OrcBoard is a really nice controller built by Edwin Olson of MIT. The Maslab team also designed a massive Java robotics API, which is very similar to my own Open Source Robot Controls project. The Maslab API is so well built that the contestants can spend nearly all their time implementing complex algorithms rather than debugging low level sensors interfaces. This is extremely important for a contest only 30 days long.
The rules are fairly simple. The robot is placed in an arena with several small red balls. The goal is to pick up as many balls as possible and place them in the scoring bins located just over the top of the arena walls. Points are earned for the number of balls inside the robot and three times the points for balls placed in the scoring bin. The full contest rules are on the Maslab website.
3d rendering of robot.My team's robot takes design elements from iRobot's Roomba vacuum cleaner. Simply making the robot circular ensures that it won't snag walls or other robots when spinning in place. We developed a front bumper element very similar to the Roomba's front bumper. It uses two touch switches and spring loaded linkages to detect if the robot hit something directly in front or to either side.
There's also a ball elevator to lift balls from the ground level and dump them on top of the robot. A motorized drum is used to divert balls into the elevator. There is a solenoid actuated gate to release the balls into the scoring bin, and a solenoid to tilt the entire top plate forward to make sure the balls roll out.
White Castle in action.Unlike many teams, who chose to start hacking in the woodshop right away, we chose to spend extra time fine tuning the design in CAD software. The modeling was done in Rhinoceros 3d, exported to DXF, and loaded into a laser cutter in the basement of MIT's Media Lab. After spending a couple solid days slaving away with the 3d model and watching the mesmerizing strokes of the laser cutter, we had a pile of white, shiny plastic parts. We called it White Castle. (Although the crown is from Burger King...) The laser cutter helped us produce a fully functional robot by the time other teams had bolted their motors to a square piece of wood. You should buy one. I think they cost $10K.
We spent the majority of our 30 days working on complex mapping and odometry systems. Unfortunately, our system was not fully functional by the contest date. Instead, we used a simple wall-following algorithm to search for red balls. When a ball was discovered, the robot first drove out from the wall and tried to pick up another ball. Next, the robot turned 360 degrees in place to look for another ball. If a ball was sensed, the robot went after the next ball. If not, it attempted to turn back to the wall and continue following it. The strategy proved to be very good at exploring the entire arena as long as there was enough time. In the end, our night-before-the-contest modifications to the ball elevator caused it to jam and actually lift the robot's wheels off the ground, stranding it. I think the robot did manage to collect a few balls and lift them to the top level before the jam.
To top of page