4.02.2004
Cars that Drive Themselves-DARPA's Grand Challenge
The Defense Advanced Research Project Agency, or DARPA (they originally funded the project that created the Internet), recently sponsored a grand challenge in which design teams from around the nation were invited to build autonomous land vehicles (cars and trucks with no human drivers) and race them in the desert between California and Nevada. The prize was one million dollars. I watched TechTV's coverage of the race and I was impressed, however, none of the entries made it further than 8 miles from the starting point. Considering all the challenges involved, I am not terribly surprised at this outcome. One hopes that they gained valuable data to help advance the concept of an autonomous land vehicle along.
Try to imagine creating an autonomous computer program (and the machinery) that can do all the things we would normally do behind the wheel of a car or truck. The permutations are endless. How will it know where it is (it's not like it can stop and ask for directions)? How will the vehicle react to people or animals in its path? What happens if the navigation system is unable to pinpoint its location? What will it do if it cannot negotiate the terrain? What will it do if it encounters an obstacle? Is it allowed to "call for help" from a human (or another machine) if it can't resolve a situation? Will it know to stop or avoid hitting humans or animals?
It seems to me that the vehicles in this race were totally reliant on GPS and had no backup system for self-correction. The designers were so focused on maintaining the GPS coordinate sets that they didn't consider providing the automated "drivers" with collision avoidance or adjustment mechanisms. It's as if the folks based their designs on the idea that the vehicles would be operating on a large empty parking lot: Perfectly flat and void of any obstacles or potential interference with their navigation systems. It's definitly a starting point, but it is obvious that there are a lot more factors to be added in as this idea evolves.
Navigation is not just about where you are on a map, but being able to correct for the unknown obstacles in the terrain as well. I want to know if it would be possible to create a sensory package using a combination of methods the way humans have five senses, so that the vehicle could "sense" obstacles and make local corrections to compensate for them, and then do an update fix on it's GPS location. My idea for a sensory package would be some form of sonar, radar, infra-red, and perhaps some sort of imaging. I guess we're looking at some way of allowing the AI to use more than just GPS as a source of navigation. Imagine if it could use star charts at night to aid in navigation, or be programmed to store key image elements of the terrain to know where it has already been.
The other thing I would want to design is some method of having the vehicle stop if it is unable to use it's sensory input or GPS fix, and run some self-diagnostics. If it cannot resolve it's own problem, it should be able to send a signal to indicate a malfunction, or request human assistance.
Some ideas I have thought of have to do with the way wireless networking works. The newest wireless standard is called 802.11g. What happens is, if you have a wireless card in your laptop that uses the "g" standard, and someone nearby (whom you've authorized) loses their direct wireless connection, they can request to "borrow" yours until their own connection becomes stable again. You'll never notice the difference. My idea would be that these autonomous ground vehicles are deployed along with aerial drones such as the Preadators, manned airplanes of various types, manned ground units, even footsoldiers, who all have some sort of electronic GPS system that the autonomous vehicle can contact to help it get a better fix on its location. I even thought that the autonomous vehicle itself could have a payload of small sensors it can "drop" that it can use as a "fix marker" of some sort. Like a tracking bug, it would be able to determine the distance from the marker and use that to better pinpoint its current location.
Probably my ideas are nothing new or original. This Grand Challenge seemed to be more about trying out vehicle configurations and basic AI to handle the simple GPS navigation issues. Perhaps as a result of this experience, more thought will be given to obstacle avoidance and redundant navigational backups, and remote sensory packages. Whatever the reasoning, it is an interesting problem to consider. Just the thing to bring out the geek in me.
DARPA Grand Challenge website
© 2004, J.S.Brown
0 comments
Try to imagine creating an autonomous computer program (and the machinery) that can do all the things we would normally do behind the wheel of a car or truck. The permutations are endless. How will it know where it is (it's not like it can stop and ask for directions)? How will the vehicle react to people or animals in its path? What happens if the navigation system is unable to pinpoint its location? What will it do if it cannot negotiate the terrain? What will it do if it encounters an obstacle? Is it allowed to "call for help" from a human (or another machine) if it can't resolve a situation? Will it know to stop or avoid hitting humans or animals?
It seems to me that the vehicles in this race were totally reliant on GPS and had no backup system for self-correction. The designers were so focused on maintaining the GPS coordinate sets that they didn't consider providing the automated "drivers" with collision avoidance or adjustment mechanisms. It's as if the folks based their designs on the idea that the vehicles would be operating on a large empty parking lot: Perfectly flat and void of any obstacles or potential interference with their navigation systems. It's definitly a starting point, but it is obvious that there are a lot more factors to be added in as this idea evolves.
Navigation is not just about where you are on a map, but being able to correct for the unknown obstacles in the terrain as well. I want to know if it would be possible to create a sensory package using a combination of methods the way humans have five senses, so that the vehicle could "sense" obstacles and make local corrections to compensate for them, and then do an update fix on it's GPS location. My idea for a sensory package would be some form of sonar, radar, infra-red, and perhaps some sort of imaging. I guess we're looking at some way of allowing the AI to use more than just GPS as a source of navigation. Imagine if it could use star charts at night to aid in navigation, or be programmed to store key image elements of the terrain to know where it has already been.
The other thing I would want to design is some method of having the vehicle stop if it is unable to use it's sensory input or GPS fix, and run some self-diagnostics. If it cannot resolve it's own problem, it should be able to send a signal to indicate a malfunction, or request human assistance.
Some ideas I have thought of have to do with the way wireless networking works. The newest wireless standard is called 802.11g. What happens is, if you have a wireless card in your laptop that uses the "g" standard, and someone nearby (whom you've authorized) loses their direct wireless connection, they can request to "borrow" yours until their own connection becomes stable again. You'll never notice the difference. My idea would be that these autonomous ground vehicles are deployed along with aerial drones such as the Preadators, manned airplanes of various types, manned ground units, even footsoldiers, who all have some sort of electronic GPS system that the autonomous vehicle can contact to help it get a better fix on its location. I even thought that the autonomous vehicle itself could have a payload of small sensors it can "drop" that it can use as a "fix marker" of some sort. Like a tracking bug, it would be able to determine the distance from the marker and use that to better pinpoint its current location.
Probably my ideas are nothing new or original. This Grand Challenge seemed to be more about trying out vehicle configurations and basic AI to handle the simple GPS navigation issues. Perhaps as a result of this experience, more thought will be given to obstacle avoidance and redundant navigational backups, and remote sensory packages. Whatever the reasoning, it is an interesting problem to consider. Just the thing to bring out the geek in me.
DARPA Grand Challenge website
TANSTAAFL!
© 2004, J.S.Brown
0 comments