Autopilot goes rogue (again)

I took the G36 up for a short flight from Watsonville to Reid-Hillview Airport in San Jose today to try to "shoot" an RNAV approach--after watching a couple of YouTube tutorials on how to do it. I set up a departure from KWVI and an RNAV approach to RWY 31 at Reid-Hillview (KRHV), in hopes that the Bonanza would actually follow the glidepath down to the runway, the way the MSFS G36 did in the video. Took off from RWY 20 at Watsonville, engaged the autopilot, set my cruising altitude, and sat back to enjoy the view as the plane climbed to 7,000 feet while flying south toward Salinas, where it looped around to fly up the Santa Clara Valley toward San Jose. Everything was going fine until I reached the first intersection of the straight-in approach to KRHV, at which point the autopilot did a U-turn and sent the Bonanza back toward Salinas. Rather than interfere, I watched to see what would the AP would do next. What it did was turn the G36 so sharply to the left that it sent the plane into a counterclockwise spin. Well, that was it. I disengaged the AP, applied right rudder to straighten the Bonanza while its nose was still down, climbed again, got back on course to Reid-Hillview, and flew the rest of the way myself. I have no idea why the AP went nuts. I hadn't touched anything until that point.

I did the flight again, but this time in a Daher TBM. I had no problems with the AP in the Daher, but the RNAV approach was a different story. When I activated the approach procedure--at the intersection marked with the little "x" on the approach plate--nothing happened. The Daher continued on toward KRHV at 2,800 feet, instead of following the glidepath down to 1,760 ft., the flight level at the next-to-last intersection. If anybody has figured out how to get RNAV to work like a virtual ILS system, I'd sure like to hear about it.