I'm sure @wk057 could or has this information but his vehicles might be AP1 right now. I mean, why the hell even mess with AP2 in its current state. I don't think Jason has improved AP1 for his personal vehicles and AP2 would be yeoman's work to get going at this point. Might as well let Tesla continue the heavy lifting. We know Jason can add AP1 to a vehicle but concluded it was a PITA. Imagine a spineectomy (HW2)?
Admittedly jumping in this thread blind, but I'll make a few notes based on the @highlight I got here.
The non-AP to AP1 retrofit I did was a pain for sure. I've looked into what it would take to do AP1 to AP2... and realistically, on a Model S, it wouldn't be nearly as bad from a hardware perspective as non-AP to AP1, assuming I had all of the components. There are some other challenges, though, since the AP2 module itself is a different beast compared to AP1. For example, I don't know if an AP2 module from a salvage vehicle could be utilized in such a retrofit. The biggest issue with an AP1 to AP2 retrofit becomes the massive amount of wiring changes required. The AP2 module sits on the ethernet bus, and the newer cars have an updated MCU. So a new MCU is needed, plus additional wiring to get the AP2 module on the ethernet bus, plus wiring for additional cameras, CAN connections, power, etc. The wiring would be the biggest hassle, assuming I could get the software to cooperate.
As for improvements to AP1... I've actually made some minor improvements.
I've made some hardware that sits in front of the AP module's CAN and the rest of the car and allows me to manipulate almost every aspect of the system. Aside from the nonsense restrictions, nags, and other things Tesla has added that take away from the quality of the system, one of my biggest issues with AP is the fact that it sees in two dimensions and thus gets super confused by lane markings in the distance when rounding a hill, causing the car to jump to one side or the other depending on perspective. I've been working on filtering these movements, mostly on a route I take frequently, as well as anticipating them in software by reading data from the car's sensors as well as logging in a small database points where the car always needs a correction, what that correction ends up being, and applying that correction based on GPS and other factors.
I've also tinkered with adding some features to the system, like actually adapting to the speed limit with an offset, as well as fully automatic highway passing. I've implemented the latter with the system warning me that it's about to initiate a lane change with a series of quick beeps so that I can check for any vehicles approaching too quickly for the ultrasonics to detect, then if I don't intervene, it will do the lane change on its own. Basically if there is an open lane to the left with a dashed line there, and it's been there for more than a mile (ie, not an on or off-ramp), and the car in front of me is going slower than my set speed, and there are no vehicles ahead in the lane to the left, or at least none going slower than the car ahead now, it will attempt an automatic lane change. It will get back over to the right after the lane is clear, also. My box tracks where the cars I'm passing should be, also, to make sure I've passed them sufficiently (camera until they leave view, and ultrasonics after that). I used it pretty successfully on my last long trip.
Another major one I've played around with implementing is a full follow-the-leader type of mode, where the car will follow a lead car completely and indefinitely through all turns, lane changes, etc, assuming it's able. Tesla's follow mode tends to break off in a full turn in either direction, but I've been able to use both map data and sensor data to extrapolate where the lead car is in a turn, and continue that turn while keeping it in view of one or more sensors... in most cases anyway. It's not perfect, and requires tricking a lot of systems to work. But it does work pretty decently for something I whipped up in less than a day. Like, if there is a lead car heading towards my house in my development (which has no lane markings) I can follow them pretty accurately, around full turns and all, right into my driveway.
I've got quite a few other things I've tweaked and other features I've been working on as well... and I've been debating with myself whether or not this is something that would be sane to market to Tesla owners. Ultimately, one of my personal goals is to be able to do hands free from my garage to some frequent local destinations... while obviously monitoring everything. Suffice it to say it's an ongoing project, but I believe I've improved my personal autopilot experience significantly beyond what Tesla's system provides.
Unfortunately, Tesla makes major changes to their CAN communications frequently, so I've had to kind of lock my car on a specific version so I don't have to constantly update things to account for their changes.
As for AP2... I've no interest in trying anything with it until it does at least some things AP1 can't. As of recently I tested out an AP2 car on newer firmware... and AP1 is still better, even out of the box without my modifications.
Last edited: