Date: Fri, 13 Jan 2006 02:02:37 -0600 (CST) Subject: stereo camera configuration and force protection X-UID: 121 Content-Type: IMAGE/JPEG; NAME="img2542.jpg" Content-Type: IMAGE/JPEG; NAME="img2543.jpg" A reader opined that half the nominal webcam framerate is sufficient. This would allow two cameras, each at half frame rate, in a stereoscopic configuration. Webcams are often specified as 30 frames per second. Due to full-speed USB limitations, performance is more like 15 fps in practice. So assuming 8 fps as the reader speculates (which is probably close to dead on) and a robot speed of 10 mph, then the robot travels 2 feet per frame. I was concerned about correlation between camera frames for stereo distance estimation. I'm not even sure the frame rates of both cameras are the same. One seems to update a little faster than the other. However, I found that even with one camera, there is noticeable image shearing. I tried both xawtv and mplayer and saw the same shearing when the camera is in motion. I think this is an artifact of sensor scanline readout and streaming. Unlike a composite video signal, the entire frame isn't updated at the same time (interlaced in that case). So when there is motion in the webcam's view, objects appear distorted. Anyway, I feel better about two cameras so am going with the stereoscopic configuration. The two cameras are laterally displaced by around 30 inches. It looks kind of weird but increases angular displacement and increases performance. The front wheels partially obscure the view of the cameras. However, it's not bad. And one advantage is the robot can visually verify wheel position. So if the potentiometer reading is inconsistent with what is seen, the robot has probably suffered damage. To mitigate damage in case of a crash, the camera tube will have a magnetic break-away mount. Six rare-earth disk magnets from the two laser line levels were salvaged. The magnets are very strong and difficult to separate when stuck together. For carrying, a rod or tube will be bolted above the robot, sort of like a carry handle. This will also provide protection in case of a rollover. It'll become clear in a few days when it's been made. I believe that all current military robot missions can be classified as reconnaissance, strike or force protection. The UAVs like Predator are used for both recon and the strike mission. It's not reported much in the media. But every few months there will be a story about a convoy or house blown apart with a missile fired from one of those machines. The DARPA Grand Challenge was about force protection through automation of vulnerable supply convoys. I know that foreign militaries use teleoperated machine guns for perimeter defense of their compounds. Anyway, this leads to where urban field robots will be first deployed. It won't be in public areas as they won't be smart enough. They will be used for guarding fortified compounds or campuses. This allows for three things. 1. proximity to a command center with human controllers 2. structure in the environment, precise maps of the compound area 3. clearer rules of engagement as it is not a public area The robot will be ready for operating this Sunday. And next week I have 24x7 oncall duty. I don't want to be caught with lots of equipment outdoors when I'm paged as the nominal response time expectation is 15 to 30 minutes before escalation. So that means the earliest the robot can have a non-trivial test outdoors is the 28th or 29th. That probably works out as I really need to write some software and do a lot of testing before doing anything outdoors. That'll take probably about a week to get something barebones put together. It's getting fun now.