Saturday, February 05, 2005
Putting my money where my mouth is - making a robot that jumps
I've been lurking in the DARPA Grand Challenge (http://www.grandchallenge.org) discussion forum since it opened, and I have to say I've been envious of those teams who are trying to create autonomous car-bots - definitely a class of Robots That Jump (if this obvious, see an off-road race). Well, the time for envy is past - I am now team leader (actually "guru") for an new GC entry, RoboMonster (http://www.robomonster.com).
The way this came about was rather remarkable. I teach web-based programming techniques at the Art Institute of California, Los Angeles, a design school. I've been promoting robotics as more than engineering - instead, an extension of interactive media design as well. Today, designers create virtual characters, website, and other intangibles. I tell the students that they might end up using the same tools to create real-world things - a real robot instead of a virtual robot, for example.
Well, one of my students has taken this to heart. In a sudden flash he decided to enter the Grand Challenge and promptly nominated me as guru. At first, I was skeptical - but I hadn't counted on his sales ability (former car salesman). In just a few days he had an incredible, custom-built "rock climber" donated to the cause, with more sponsorships in the pipe. So it's official - we're going to try to qualify for the 2005 Challenge, though I suspect the real race (with a winner) will be in 2006.
One of the reasons for my interest is the vehicle. The GC is a fairly rough desert course with unpaved roads, rocky areas, etc. Several GC teams have driven the 2004 course themselves and come back with their teeth practically rattled out of their sockets. An SUV-based GC entry can make it, but it will require serious robot smarts. The larger vehicles (like CMUs Humvees) actually aren't that much better. The best choices are desert-racing vehicles. Some teams (e.g. top-seeded Cyberrider) have modified dune buggies. But these vehicles are relatively low to the ground, requiring that the sensors be mounted on a high platform.
In contrast, a rock-climber has huge wheels and an engine/transmission creating huge torque - specifically for going over fields with rocks the size of a person. It has the relative lightness and desert-design of a dune buggy, but may be tougher when encountering obstacles not detected by the robot sensors. The particular model we have has a very smooth power steering and transmission, ideal for converting to drive-by-wire.
The plan is to provide autonomy along a redundant, sensor-loaded model with "bottom up" behavior-based programming. In particular, we plan to put a much higher density of sensors on the robot (e.g., 20 cameras) on the principle that robots improve according to the rate of doubling sensor density I have discussed in other postings. These sensors will be connected to a dozen or so single-board computers/DSPs to build simple sensor maps. Arbitration will follow behavior-based programming principles.
Another feature of our entry will be style. After all the team leaders come from a media design school! So expect plenty of snaz to the look, including active displays, animatronic components, end a bio-morphic behavior. My doctorate is in biology, so I'm already planning how to design biologically-inspired features of RoboMonster to make it seem like a big critter rather than a Hal 9000 disembodied intellect.
So, if you're out there and would like to contribute, drop a line to pindiespace@robomonster.com or pindiespace@gmail.com. We're currently looking for a kick-ass embedded systems programmer to compliment the mechanical/electrical skills of the team, as well as game engine programmers. Wish us luck!
I've been lurking in the DARPA Grand Challenge (http://www.grandchallenge.org) discussion forum since it opened, and I have to say I've been envious of those teams who are trying to create autonomous car-bots - definitely a class of Robots That Jump (if this obvious, see an off-road race). Well, the time for envy is past - I am now team leader (actually "guru") for an new GC entry, RoboMonster (http://www.robomonster.com).
The way this came about was rather remarkable. I teach web-based programming techniques at the Art Institute of California, Los Angeles, a design school. I've been promoting robotics as more than engineering - instead, an extension of interactive media design as well. Today, designers create virtual characters, website, and other intangibles. I tell the students that they might end up using the same tools to create real-world things - a real robot instead of a virtual robot, for example.
Well, one of my students has taken this to heart. In a sudden flash he decided to enter the Grand Challenge and promptly nominated me as guru. At first, I was skeptical - but I hadn't counted on his sales ability (former car salesman). In just a few days he had an incredible, custom-built "rock climber" donated to the cause, with more sponsorships in the pipe. So it's official - we're going to try to qualify for the 2005 Challenge, though I suspect the real race (with a winner) will be in 2006.
One of the reasons for my interest is the vehicle. The GC is a fairly rough desert course with unpaved roads, rocky areas, etc. Several GC teams have driven the 2004 course themselves and come back with their teeth practically rattled out of their sockets. An SUV-based GC entry can make it, but it will require serious robot smarts. The larger vehicles (like CMUs Humvees) actually aren't that much better. The best choices are desert-racing vehicles. Some teams (e.g. top-seeded Cyberrider) have modified dune buggies. But these vehicles are relatively low to the ground, requiring that the sensors be mounted on a high platform.
In contrast, a rock-climber has huge wheels and an engine/transmission creating huge torque - specifically for going over fields with rocks the size of a person. It has the relative lightness and desert-design of a dune buggy, but may be tougher when encountering obstacles not detected by the robot sensors. The particular model we have has a very smooth power steering and transmission, ideal for converting to drive-by-wire.
The plan is to provide autonomy along a redundant, sensor-loaded model with "bottom up" behavior-based programming. In particular, we plan to put a much higher density of sensors on the robot (e.g., 20 cameras) on the principle that robots improve according to the rate of doubling sensor density I have discussed in other postings. These sensors will be connected to a dozen or so single-board computers/DSPs to build simple sensor maps. Arbitration will follow behavior-based programming principles.
Another feature of our entry will be style. After all the team leaders come from a media design school! So expect plenty of snaz to the look, including active displays, animatronic components, end a bio-morphic behavior. My doctorate is in biology, so I'm already planning how to design biologically-inspired features of RoboMonster to make it seem like a big critter rather than a Hal 9000 disembodied intellect.
So, if you're out there and would like to contribute, drop a line to pindiespace@robomonster.com or pindiespace@gmail.com. We're currently looking for a kick-ass embedded systems programmer to compliment the mechanical/electrical skills of the team, as well as game engine programmers. Wish us luck!
