MASLab 2004 Contest

Eighteen teams participated in the fourth annual MASLab Robotics Exhibition and Contest which took place on January 30, 2004. Over 500 spectators crowded into room 26-100 to watch the students show off their hard work and they were not disappointed. Many robots successfully collected several red balls and then deposited then either in front of or inside the mouse-holes.

Media

Photographs of the competition are in the MASLab 2004 gallery.

Contest Rules

(C) Jonathan Wang 2004

Robots will be placed in a playing field whose layout is completely unknown. Students have seen the components of the playing field before, but they do not know the shape or size of the field until the robot is switched on! Robots must then explore the playing field gathering targets (red wooden 2.25" balls), and then deliver them to scoring areas. Scoring areas are actually 6"x10" mouse-holes in the walls. Teams can score three points by leaving targets in front of the mouse-hole or they can score five points by maneuvering targets through the mouse-hole. Teams also score one point for each non-scoring target in the robot's possession at the end of the round.

In addition to free targets, there will also be green towers on the playing field. These 7.5" tubes always contain exactly three red wooden balls. Teams can move the tubes or knock them over. Tubes are not used for scoring (although the red balls within the tubes are definitely used for scoring). The official contest rules specify that there will be one to six more red balls in towers in comparison to balls left free on the playing field.

The contest is design to allow a wide variety of solutions with a wide range of implementation difficulty. Those students with strong backgrounds in mechanical engineering, for example, can score more points by performing more complicated tasks with intricate gripper mechanisms. Students from an AI background can score more points by developing algorithms which allow them to efficiently explore the playing field, making better use of a simpler gripper.

Strategy

Each team's robot is a unique creation with its own mix of mechanical, electrical, image processing, and high-level strategies. MASLab robots involve considerable mechanical design: each piece of the robot must be carefully positioned for the robot to remain within the maximum dimensions and a gripping mechanism is required to transport the targets to scoring areas. The robots in this year's contest approached the problem in many different ways. Some robots used simple gated enclosures to collect red balls and added special mechanisms for tipping green towers. Other robots ignored the red balls and used forklifts to lift the green towers and move them to the scoring areas. Several robots constructed two finger grippers which pivoted to lift balls, towers, or both off the ground. Other robots made use of rollers to sweep balls into an internal enclosure. One robot had a very unique contraption which used an elevator to move balls up a PVC tube before rolling them down a ramp and into the mouse-hole.

Each team had to make various design trade-offs when working on the electrical aspects of their robot. Some teams used short range infrared sensors or simple momentary switches to detect when their robot collided with an obstacle, while other teams used spikes in their robot's motor currents to determine when their robot was in trouble. Teams also made use of long range infrared and ultrasound range finders to look for open space and create more efficient exploration algorithms. Optical encoders allowed teams to move more precisely by giving feedback on how fast their wheels were actually turning.

Teams spent a significant amount of time working on their robot's image processing. Teams used color segmentation, edge detection, iterative sub-sampling, and feature recognition to find targets and scoring areas on the playing field. Some teams opted for simpler algorithms to increase their robot's frame rates and thus the speed at which their robots could travel, while other teams used more sophisticated algorithms to increase their precision but at the expense of a slower robot.

Finally, teams had to integrate the mechanical, electrical, and image processing aspects of their robot into a cohesive high-level strategy. Some teams focused only on the red balls or the green towers, while other teams tried to handle both types of objects. Teams used a simple timer to decide on various robot behaviors (i.e. when to start looking for scoring areas) or they used used a more sophisticated priority based scheme to decide what their robot should do next. Robots included exploration algorithms to discover targets that were in remote parts of the playing field, and these exploration algorithms included: simple random walks, wall following, wall gap detection, and open space wandering. While four minutes sounds like a long time, a robot can easily squander it by spending too much time collecting and processing sensor data, or by picking an inefficient route from one point to another. Acquiring more data enables the robot to choose more efficient routes, so the time limit imposed a fundamental design trade-off.