sensoarltd
commented
2 years ago We have also tried reconciliation of logs from the SIM and DAA hardware to match sim events with hardware detections. We had around a 3 second delay due to stream delays and the slower RT performance of the pi verses SITL on the workstation. This would mean the detections were always delayed by too long to extract meaningful detection distances.
No solutions have been found currently, a new approach may be of merit embeding sim timestamps with the video stream or playing backthe simulation recording on the DAA hardware to eliminate stream render lag.
We have tried signaling DAA detections running on real hardware back to the simulator script but there is considerable lag of up to 1.5s using a Mavlink approach. A seperate socket to listen to the DAA hardware is a possible fix. In version 1.0 the approach we used was to compare log files running on the simulator and on the DAA hardware.
One issue with the current approach is that we could not detect false alarms.