Source code / additional notes for the current test platform can be viewed here.
Antweight build for Robothon 2007 - it's done! (click to enlarge)
Antweight build for Robothon 2007 - about 80% done (click to enlarge)
New 9oz test platform (click to enlarge / see part labels)
The basic concept is that it should be possible to control the direction of a thwack-bot (a bot where the entire body is spun by the drive train for use as a weapon) if an onboard computer brakes or accelerates each drive-train motor at certain points during each spin at just the right time. A human driver still controls the robot - but the computer is handling direct control of the drive motors.
The entire mass of the robot could be used as a weapon, and actively driven towards the opponent while spinning. In theory, this could result in a uniquely deadly bot.
The current version of Melty B is a beetle-weight (3lbs).
Melty B 1.0 being used as a test platform for future development
Melty B made its combat debut at Robothon 2006. It encountered a number of problems including underpowered batteries and a body design that got caught on the arena floor. In short, the bot "wasn't quite there" yet.
Team Death by Monkeys has video of the event including one of my not-so-successful fights.
While Melty B did not perform well in combat, it did work well enough to put on a somewhat-convincing display of translation drift. As a result, it won the Robothon Design Award.
Melty B's mainboard (if wire wrap was good enough for the Apollo guidance computer...)
Many builders have attempted to create this kind of robot - to varying degrees of success. Cyclone may be the best known. Many undoubtedly have worked well in test-environments, but like Melty B, hit problems when trying to actually fight another robot.
This class of combat robots is still experimental. To my knowledge, no one out there is consistently winning fights.
There are several obstacles to creating a melty brained robot. In my opinion, the most difficult one is that the robot must "know" what direction it is facing at any given point in time.
Melty B's approach is to use an accelerometer to accurately determine the spin rate of the bot. It flashes an LED each time it hits a point in the spin it thinks is "forward." The effect is that the LED appears as the front of the blurry spinning mass which is the robot. The driver can hopefully compensate for any error via the remote.
This works moderately well in a test environment. However, there is still some "drift" - especially if the bot comes into contact with another object. Future versions of Melty B may use an external infrared beacon to provide a common reference point; hopefully eliminating this problem.
Melty B 1.0 uses the following (amongst other) components:
Atmega 168 Microcontroller (on a Baby Orangutan board)
Pololu's High-Current Motor Drivers
Dual Axis Accelerometer +/-18g from Sparkfun