Thread Tools
This thread is privately moderated by Jack Crossfire, who may elect to delete unwanted replies.
Mar 05, 2016, 12:21 AM
Registered User
Jack Crossfire's Avatar
Thread OP
Discussion

7m15s mile


7m15s mile (8 min 4 sec)


Snagged a 7m15s mile from the Ruckus. The image stabilization had its 1st material improvement in 18 years. There were previous attempts, but nothing which gave a net improvement. Now, much higher resolution can be stabilized in translation & rotation in much less time. This one took only 3 hours on the Macbook. On the dinosaur PC, it took 5 hours. Previously, the dinosaur PC was slightly faster than the Macbook.

The mane focus was using mipmaps instead of the full resolution image. The original algorithm stressed memory access while the new algorithm stresses the cache more. It shows the Macbook may have slower memory bandwidth, but faster cores. Memory may only be slowed down by the virtual machine.

There's still no solution to rotation drifting. Stabilizing a flipped video showed it's not a rounding error.

Then came the reality that using the vehicle as a pacer over a full 2 miles resulted in a slower total time than running without it, because of all the flips & navigation through mobs. The vehicle still allows higher speeds when maneuvering isn't required. Surprising how much a phone screen scratches after many slides on the pavement.

Steering oscillation remaned pretty bad. Steering was also too slow to respond, symptomatic of high D constants to overcome the oscillation. These factors made it very hard to navigate obstacles.
Sign up now
to remove ads between posts
Mar 05, 2016, 01:29 PM
jsr
jsr
Registered User
Re: steering oscillation. Are the gains used on the Ruckus steering constant, or do you schedule them with some other parameter such as speed? Gain scheduling can be a nice way to keep a system stable and high performance in varying conditions, like over a wide speed range.


Quick Reply
Message:
Thread Tools