Tesla has started rolling out a big new beta update to its self-driving software (FSD) with many improvements.
According to CEO Elon Musk, the automaker may expand access to the beta if this update “shows good results.”
Full Tesla Autonomous Driving Beta
Since October 2020, Tesla has been slowly rolling out the so-called “Full Autonomous Driving Beta” (FSD Beta), which is an early version of its self-driving software currently being tested by a group of Tesla owners selected by the company and through its “assessment security.”
The software allows the vehicle to autonomously navigate to a destination entered into the vehicle’s navigation system, but the driver must remain alert and ready to take control at any time.
Since the responsibility lies with the driver and not with the Tesla system, it is still considered a second level driver assistance system, despite its name. It was sort of a “two steps forward, one step back” type of program, as some updates saw a regression in terms of driving capabilities.
Tesla frequently releases new software updates to the FSD beta program and adds new owners to it.
The last major update was FSD Beta 10.10 in early February.
As of Q4 2021, the automaker said it had almost 60,000 owners in the FSD beta program.
Tesla FSD Beta 10.11
Now Tesla has started rolling out the new FSD Beta 10.11 update to its early access program, and judging by the release notes, it’s an important one.
Here are the release notes:
- Advanced track geometry modeling from dense rasters (“bag of dots”) to an autoregressive decoder that directly predicts and connects “vector space” stripes point by point using a transducer neural network. This allows us to predict band crossings, provides computationally cheaper and less error-prone post-processing, and paves the way for predicting many other signals and their relationships end-to-end. Use more accurate predictions of where vehicles turn or merge to reduce unnecessary deceleration for vehicles that do not cross our path.
- Improved understanding of the right-of-way if the map is inaccurate or the car cannot follow the navigator. In particular, intersection extent modeling is now entirely based on network predictions and no longer uses map-based heuristics.
- Improved VRU detection accuracy by 44.9%, significantly reducing false positives for pedestrians and cyclists (especially around bituminous seams, skid marks and raindrops). This was achieved by increasing the data size of the next generation autolabeler, learning network parameters that were previously frozen, and modifying network loss functions. We found that this reduces the frequency of false slowdowns associated with VRU.
- Reduced predicted speed error for very close motorcycles, scooters, wheelchairs and pedestrians by 63.6%. To do this, we have introduced a new dataset simulating high-speed enemy VRU interactions. This update improves autopilot handling around fast moving and crashing VRUs.
- Improved creep profile with higher jerk at the beginning and end of the creep.
- Better control of nearby obstacles by predicting a continuous distance to static geometry using a common static obstacle network.
- 17% reduction in the error rate of the “parked” vehicle attribute due to a 14% increase in data set size. The accuracy of stoplights has also been improved.
- Reduced the speed error in the “clean path” scenario by 5% and the speed error in the “highway” scenario by 10% by tweaking the loss function to improve performance in complex scenarios.
- Improved detection and control of open car doors.
- Cornering smoothness has been improved by using an optimization-based approach to determine which road lines are irrelevant to control, taking into account lateral and longitudinal acceleration and jerk limits, as well as vehicle kinematics.
- Improved stability of FSD Ul visualizations by optimizing the Ethernet data transfer pipeline by 15%.
CEO Elon Musk says that if this update “works well,” Tesla will “probably” reduce access to drivers who score “95” on driver safety. This should give a lot more FSD buyers access to the beta.
In Canada, Tesla owners who bought an FSD have had access to a driver safety assessment for more than a week now. The automaker is expected to soon start releasing a beta version for those who have achieved the highest scores.
FTC: We use automated affiliate links that generate income. More.
Subscribe to Electrek on YouTube to receive exclusive videos and subscribe to the podcast.