February 11, 2004

Field Test Report- February 11, 2004

Objectives

Test with new DMI
Verify ability to run independently recorded paths coming from the buffalo DGPS
Drive the Hootin Wells dirt road

Report

The spare DMI was installed and confirmed to work. Two laps of the dirt oval were run cleanly and without incident. Sandstorm was then paused for a period of time. On the third lap, Sandstorm drove around the first 1/4 of the loop before its apparent state shifted such that it was half in the ditch around the oval. There appears to be some correlation between Sandstorm sitting still and drift in it's measured location. A theory is that the DMI is generating spurious pulses which are confusing the applanix's state estimation code. Debugging continues. We are developing a work around solution to replace the half shaft mounted encoder with an encoder mounted on the output of the transfer case. More to follow.
The laps Sansdstorm ran around the dirt oval were recorded with the buffalo DGPS. Sandstorm followed the path cleanly. It's possible that an overly high sampling rate for points on the path induced some oversteer in the path tracker. The test was successful.
Sandstorm was loaded onto an open trailer and driven the 5 miles to the Hootin Well's trail head. For the outbound leg, Sandstorm was commanded to drive at a steady speed of 10m/s (22mph), slowing down only as appropriate for cornering. Sandstorm drove the 16 miles of the outbound leg without incident. 2 stops were made on the way out, the first was to study the behavior of the intelligent gimbal lookahead code, which proved to be working correctly. The second stop was to transfer through cattle gates. On the return leg, Sandstorm drove at a commanded velocity of 15 m/s (33mph) successfully, with only one stop to be let through the cattle gates.

Observations

Velocity control appeared to be quite smooth even when traveling over significant hills.
Today we saw Sandstorm's first traverse of a dirt road that many of us hadn't seen until Sandstorm drove it autonomously. The data was provided from a DGPS receiver other than the applanix- another first.Things are starting to come together.
There is something surreal about driving through a giant dust cloud at 30mph with only the bulk of the driverless robot in front of you as a guide to where the road is.
The Scientific American article about the Redteam came out today- it's quite excellent. We're also mentioned as a front runner in Esquire magazine.

Photos & Videos

Sandstorm trucking across the desert

Sandstorm continues to move

Sandstorm and its entourage of the day

Sandstorm moving at 33mph

Sandstorm ascends a hill climbing out of a dry lakebed.

A graph showing Sandstorm's velocity control response

The velocity control image plots desired speed (red line) and actual speed (green line) (in m/s- multiply by 2.2 to get mph) vs. control step (the control loop runs at 10 hz, so divide by 10 to get seconds). Note that the actual velocity almost always stays within 0.5 m/s of the requested speed, the slightly larger deviation occurs while Sandstorm climbed and descended a significant hill. As an observer in the chase vehicle, Sandstorm's descent appeared controlled and it was not apparent that it was accelerating or oscillating around the commanded velocity. The error in the control loop has an apparent frequency of 0.1 Hz.

Videos
Sandstorm in motion- (3 mb wmv)