Author / Aaron
Tesla has been slowly launching the FSD Beta version since October 2020 and has been tested by a group of owners of its pick. Tesla has said that all owners who have access to the FSD beta version must pass the assessment of the "safety scoring system", which is measured by the need for their driving behavior to perform well for 7 consecutive days and reach more than 98 points.
It is understood that the "safety scoring system" is a scoring system introduced by Tesla last year to evaluate the safety behavior of drivers, through every 1000mile of forward collision warning, emergency braking, sharp turns, unsafe follow-up, forced release of Autopilot five indicators, to judge the owner's driving habits.
Under U.S. law, the responsibility for an accident in a current self-driving car rests with the driver and not with the autopilot system. Moreover, although Tesla named the FSD (Full Self-Driv) a fully autonomous driving system, its capabilities are still at the L2 level. The market believes that this is a way to take two steps forward and take a step back, because although FSD Beta will often update or release some new functions, its autonomous driving system's ability to respond autonomously to new complex scenarios has not only not improved compared with the past, but has seen signs of regression.
As of the fourth quarter of 2021, the company said that nearly 60,000 car owners had participated in the FSD Beta program, and the most recent major update was FSD Beta 10.10 in early February 2022, but the version did not work well.
Tesla FSD Beta 10.11
At present, Tesla has begun to promote a new FSD Beta 10.11 version, which the official said is an extremely important update. Tesla CEO Elon Musk said that if the update "performs well", Tesla is likely to lower the entry criteria for participating in the FSD beta version, expanding the push scope to vehicles with a safety driving score of 95.
According to Musk on Twitter, the vector lanes in FSD Beta 10.11 are a major architectural improvement to Tesla's ARTIFICIAL. This will enable vehicles to more accurately predict cross lanes, reducing unnecessary deceleration when turning and merging.
In addition, according to the release notes of FSD Beta 10.1, this version also makes certain fixes to the problems encountered in the 10.1 version, and the functions and scene performance are more perfect. Referring to a video of FSD Beta 10.10 uploaded by YouTuber user AI Addict in February this year, the system has the following series of problems due to ability limitations:
Drove into the tram tracks;
Crashing into a bike lane bollard at 11 mph (about 17.7 km/h);
No stopping give way in front of the zebra crossing where pedestrians are about to pass;
Long-distance parking;
Fight for control of the steering wheel;
Some traffic signs are not recognizable.
On this update of the FSD Beta 10.11 version, Tesla reduced the vehicle parking error rate by 17% by increasing the size of the dataset by 14%, and also improved the accuracy of braking timing. At the same time, in the case of inaccurate maps or cars cannot follow navigation, the FSD algorithm can also improve the prediction of roads and improve the understanding of road rights.
FSD Beta 10.11 also features Tesla's next-generation auto-labeling tool, which improves detection rates for vulnerable road users (VRUs) and reduces false positives for "cyclists and pedestrians" by 44.9 percent, which is a problem that plagued the previous version.
Here are the specific updates:
Modeling lane geometry is upgraded from dense rasters (dot bags) to an autoregressive decoder that uses a transformer neural network to directly predict and connect "vector space" lanes point by point. This enables vehicles to predict cross-lanes, i.e. post-processing that allows for lower computational costs and fewer errors, and paves the way for predicting other signals and their joint and end-to-end relationships. Use more accurate predictive vehicle turns or parallel algorithms to reduce unnecessary slowdowns.
If the map is inaccurate or the car cannot follow the navigation, the FSD algorithm can further improve the understanding of the right of way on the road. In particular, the modeling of intersectional segments is now based entirely on network predictions and no longer uses map-based heuristic models.
The VRU detection accuracy was improved by 44.9%, significantly reducing misjudgments about fake pedestrians and bicycles (especially near asphalt cracks, brake marks and raindrops). This is achieved by increasing the data size of the next generation of auto-annotation tools, freezing network parameters before training, and modifying the network loss function. Overall, the incidence of false deceleration associated with VRU was reduced.
Reduced the predictive speed error of approaching motorcycles, scooters, wheelchairs and pedestrians by 63.6%. To this end, FSD introduced a new simulated adversarial high-speed VRU interaction dataset.
Improved Creeping Mode, which now has a higher variable acceleration at the beginning and end of idle.
Control of nearby obstacles is enhanced by predicting static geometry of continuous distances and the general static barrier network.
By increasing the dataset size by 14%, the vehicle parking error rate was reduced by 17%, and the accuracy of the brake lights was also improved.
By adjusting the loss function, the performance in many difficult scenarios was improved, and the speed error of the clear scene was increased by 5%, and the speed error of the road scene was increased by 10%.
Improved detection and control of open doors.
By optimizing methods, which routes do not require control, given lateral and longitudinal acceleration and acceleration limits, as well as vehicle kinematics, improves the smoothness of the turns.
By optimizing the Ethernet data transmission pipeline by 15%, the stability of the FSD Ul visualization is improved.
About Auto Byte
Auto Byte launches the Automotive Technology Vertical Media for The Heart of Machines, focusing on cutting-edge research and technology applications in the fields of autonomous driving, new energy, chips, software, automobile manufacturing and intelligent transportation, and helping professional practitioners and relevant users in the automotive field understand technology development and industry trends through technology to gain insight into products, companies and industries.
Welcome to follow the stars and click Likes and Likes and Are Watching in the bottom right corner.