Accomplished
This week I filmed longer squat-rack videos and tested them offline on the Jetson before moving the workflow to a live camera feed. Our earlier clips were useful for checking individual states, but the longer recordings gave me more continuous activity to review and repeat while testing.
After the offline tests, I got the rack workflow running directly from the camera on the Jetson. It now captures and processes incoming frames rather than reading from a saved video. I focused on whether inference keeps up with the feed and whether the observations used to update the rack state are still current.
I also began profiling memory usage, per-check latency, and throughput during these runs. These measurements give us an individual baseline before we combine the rack workflow with the treadmill and bench checks. Live inference is now running, but the full camera-to-tablet path and five-minute unattended scenario still need testing.
Schedule status
The live camera feed completes the next deployment step after last week’s offline Jetson run. Profiling is underway, but we still have not completed a combined three-workflow test or connected the full live path to the tablet, so integration remains behind our earlier schedule.
Next week I will prioritize the shared Jetson run with Jerry and Alan. I will compare the rack’s individual measurements with its performance under the combined workload and help address any processing delays before the interim demo.
Deliverables for next week
- Run the live rack workflow alongside the treadmill and bench workflows on one Jetson.
- Compare individual and combined memory usage, latency, and throughput.
- Connect live rack observations to the tablet display.
- Test a full five-minute unattended scenario and a return before the timer expires.