10/25/2017: While testing the trajectory client on the Lo-Co, all terminals opened in the Jetson stopped receiving keystrokes. We discovered shortly after the Jetson was no longer powered. Furthermore, pressing the power button did nothing and the Red LED (what does this mean?) was dark. We removed the battery and it showed a voltage of 4.8V across all three cells.
Since the Jetson should shut off at ~11 V, we believe we shorted something in the electronics that caused the battery to discharge quickly. We are planning on doing a thorough check with the multimeter to determine if there are any blown fuses or shorts.
11/7/2017: While attempting to power the Lo-Co vehicle from the wall, a team member (who will remain unnamed) applied a 19V power source to the + end of the power distribution board to smoke. Initially this caused a loud noise and the power distribution board began to smoke.
Upon further investigation, the reverse voltage protection of the circuit worked. The reverse voltage diode created a short to ground rather than forcing current the wrong way through the DC-DC converter. A new diode had to be found and re-soldered in, at which point the power distribution board and vehicle worked as normal once again.
11/28/2017: PDB doesn’t have a connection of one test point to the line. Connection needs to be added in the new design.
02/14/2018: There is a lag in the tracking of pedestrians using the Jetson TX2. It is possible that this is just a result of some mathematical errors, but we should look closer at whether there is something in the detection slowing down the system.
02/06/2018: Training the Social LSTM without any skips for the interpolated dataset results in a non-positive Covariance matrix which could either have been a result of numerical instability (correlation value being very close to 0) or because there is a high correlation between the rows of the Covariance matrix. Chances of this being true is very small, and anyways this happens after we get the best validation loss so we don’t need to worry about it too much.
03/12/2018: Testing with the Jetson TX2 in real time is proving problematic because to store a bag file – the write speed to a hard drive is not giving consistent frames. To partially solve this problem we have decided to record bag files from the desktop and then run on the Jetson. We have also purchased an SSD which should give a much faster write speed from the TX2.
03/19/2018: The GPS module is able to connect to the internet but we are unable to reconnect to the GPS module in order to configure it with other WiFi networks (This issue still persists as of 04/03)
03/21/2018: The transceiver that was purchased came with difficult to use cables and hence we have fabricated our own since this is a cable that will have to be handled frequently.
04/11/2018: Second Velodyne stopped transmitting data via ethernet because of an internal wire getting damaged. We fixed this by extracting the damaged wire and soldering it as a stop-gap measure.
04/17/2018: The fusion algorithm is slow because of latency issues with the image perception pipeline. Using compressed images instead of raw images has mitigated this issue somewhat.
04/23/2018: The weather is causing some difficulty with testing. We have purchased the DeWalt power hub to help increase the length of time that we can test outdoors. We also purchased a canopy to keep our infrastructure protected from the rain and wind while testing.