Date: Wed, 3 May 2006 04:00:42 -0500 (CDT)
Subject: monocular road edge estimation
X-UID: 169

http://golem5.org/robot1/video/drive02.mpg

I didn't really get any work done the last week. It was my turn to handle 
24x7 production support oncall during a bad week. So I ended up spending 
about two extra days in addition to the normal five for the day job. It 
may also turn out that I have to continue for another week as the next 
person is indisposed. If this happens, due to the rotation schedule, I may 
have four months free of the pager duty. That means more time for robot 
development.

Anyway, I made sure to get out of work yesterday after eight hours and 
drive the robot in a new place. Finally, there is code that demonstrates 
the monocular distance estimation idea. It kind of works. Shadows, ground 
clutter, lens flare, these all confuse the algorithm. However, the 
algorithm does seem to otherwise handle the specular component of 
reflected sunlight off the road. What is encouraging is that there is 
almost no code and it only took a few hours to figure this out.

It is becoming clear that vision requires quite a bit of intelligence in 
the manner of clever algorithms. Finding these solutions is not easy as 
they are not obvious. Just as with everything else about the robot, it is 
driven by a combination of guesswork and experiment.

Driving seems to work but handling is still tricky. The robot has a 
personality of it's own sometimes. This bothers me. Once the code was 
redesigned to include record acquired data from the sensors and monocular 
camera, the "buttery smoothness" disappeared. I don't know if this 
indicates that something is fundamentally wrong.

A person posted to the DPRG that DARPA has officially announced the next 
robot race. This one will be through an urban environment. It is scheduled 
for late 2007. It is a 60 mile race through a cluttered environment with 
moving vehicles. Robots must obey traffic laws. The signal processing is 
probably simplified from the desert race due to structure in the urban 
environment. But the level of reasoning intelligence required is much 
higher.