Tesla Model S owner and blogger KmanAuto recently released video from an Autopilot/Autosteer test (using firmware 8.xx) on winding two-lane road.
While these types of roads are not what Autosteer was developed for, it is still surprising that 8.xx performs worse than 7.xx.
According to a comparison between 7.xx and 8.xx on the same road (8 video above, older 7 below), the new firmware struggles more to keep car on track.
We assume that Tesla has focused its priorities, and use of sensors, to improve Autopilot on major roads and highways.
The side effect of that effort, it would seem, are worse results in other areas/types of driving situations.
To achieve better results Tesla probably will simply need to add more scanning equipment on-board.
Tesla Motors: Wooded Road Autopilot Test Low Speed Firmware 8.xx
================================================= An update to a previous test on the same road. I did this test on firmware 7.xx about 6 months ago. Now, with Firmware 8 and it's claimed improvements, I did it again, and it did not fare too well. In my opinion, as the driver, it did worse then previously on the 7.xx firmware's. You can see the Firmware 7.xx Attempt at https://www.youtube.com/watch?v=S-LP0...
Personally, firmware 8 has been a train wreck of bugs and regressions. Anyways, judge for yourself. Conditions were fairly similar between both tests.
My Quote on the Autopilot Performance in Firmware 8: "In general, some improvements have been made, yet, also some major regressions. One major improvement is the "X-Ray" Radar, that is able to see ahead of the vehicle in front of you. This seems to have sped up reaction times to sudden braking and stops. Especially prominent in Close quarters stop and go traffic, or when approaching a standstill on the freeway while at speed. However, I have noticed regressions in performance as well. On roads I travel excessively, and which previous autopilot worked flawless (or as flawless as one could hope for the current hardware suite), the vehicle is now having considerable problems keeping in its lane, despite a good positive hard lock on either one or both lanes. It wonders and "pings" back and fourth considerably.
Though, it performs like it did previously when following behind another vehicle which it can also clearly see.
Rain performance is a no go. I have shown some awesome autopilot videos in downpours previously, where as in much better conditions I cannot even get a lane or vehicle lock. Roads where lanes split (Ex, right lane makes a right break away, while the left lane continues straight on free way), if I'm traveling in the break away lane, with a good solid lock on both lines, instead of following my lane (which both lane lines beer right) the car will continue traveling straight right through the line (and almost into concrete barrier in some cases). While I know autopilot was not designed for the road in this video, it provides a safer, worst case scenario road, that I'm able to mostly recreate conditions. Sun, weather, lines, traffic, speeds). Testing on this road, I know it will generally "fail" in its ability to fully navigate me, however, it also allows me to show how each version of the firmware and autopilot (and in the future, full autonomous) can handle the same failure situations. also weather one version is more failure prone in comparison to others.
or improved and handle additional situations that previous (or possibly future) versions cannot."