Skip to main content

2 docs tagged with "benchmark"

View all tags

Does command timing show up as arm shake? We measured a 6 Hz vibration in two phone videos of a humanoid

In our previous post we moved a humanoid's arm and torso commands onto the WIM real-time stack and measured their timing on the CAN bus. This time we measured the fast vibration of the arm in two phone videos of the same robot, using point tracking. On the stock control stack the whole arm vibrated together at about 6 Hz (correlation between points on the arm 0.99). On the WIM real-time stack that vibration was not there (0.11). The vibration was about 10 times smaller in screen pixels, and we estimate 3 to 5 times smaller after correcting for camera distance.

Sending a humanoid's arm commands exactly every 4 ms: a stock control stack vs a real-time stack under 10 conditions

We moved the arm and torso commands of a wheeled dual-arm humanoid from its stock control stack to the WIM real-time stack on a Jetson AGX Orin, which writes them to CAN directly. Under 10 conditions with the same motion, the stock stack showed a p99 jitter (deviation of the command interval from 4 ms) of 4.0–21.9 ms even with the same CPU isolation and real-time priority applied, while the WIM stack held 0.13–0.41 ms. In a 30-minute comparison run, intervals longer than two periods occurred 18,422 times versus 0.