Tesla has released a new version of FSD Beta 10.5 that includes some fascinating new features. Read here.
Monday, November 22, 2021 | Chimniii Desk
Key Highlights
- This weekend, Tesla began rolling out a new Full Self-Driving Beta software update (10.5) to the fleet.
- This is an early version of the firm's self-driving software that is being tested by a fleet of Tesla owners chosen by the company based on their "safety test score.
- Tesla increased static world detection by "up to 13%" by installing a new auto labelling system that utilised 165,000 fresh movies.
- The automaker has been creating technology to automatically label content for machine learning, and it appears as though it recently installed one that assisted in training this new FSD version 10.5.Another feature that many people noticed was the inclusion of "emergency collision avoidance manoeuvres" in shadow mode.
- This new Full Self-Driving Beta 10.5 software is scheduled to be evaluated for the next few weeks before Tesla releases a new version, which should allow more owners who paid for it to use the software.
Advertisement
This weekend, Tesla began rolling out a new Full Self-Driving Beta software update (10.5) to the fleet.
According to the release notes, which are included below, it has several fascinating new features.
Tesla is now rolling out what the business is calling Full Self-Driving Beta (FSD Beta). This is an early version of the firm's self-driving software that is being tested by a fleet of Tesla owners chosen by the company based on their "safety test score."
Although the software permits the vehicle to travel independently to a destination specified in the navigation system, the driver must always be aware and prepared to take control.
Due to the fact that the driver, not Tesla's technology, bears responsibility, despite its name, it is still regarded a level 2 driver-assist system. It's been a bit of a "two steps forward, one step back" type of programme, with certain updates resulting in regressions in driving skills.
In previous upgrades, the automaker has included additional details regarding the driving behaviour modifications in the release notes.
That is once again the case with a new update (10.5) that Tesla began pushing to owners in the US Beta programme who have a safety score of 98 or higher.
Advertisement
Tesla noted the following changes in version 10.5's release notes:
- Enhanced the velocity error of VRU (pedestrians, bikers, and motorcycles) crossings by 20% as a result of improved quality in our auto-labeling.
- Improved static world predictions (road lines, edges, and lane connections) by up to 13% through the addition of 165K auto-labeled films and a new static world auto-labeler.
- Increased the precision and recall of cone and sign detections by augmenting the generalised static object network with 15K more video clips and altering oversampling and overweighting algorithms (+4.5% precision, +10.40% recall).
- 5.5 percent improvement in the cut-in detection network to help eliminate erroneous slowdowns.
- In shadow mode, enabled "emergency collision avoidance manoeuvring."
- When it is safe to do so, enabled the behaviour to lane change away from merging.
- Recall of merging objects improved with the use of multi-modal object prediction at intersections.
- Improved merge control by enhancing the smoothness of arrival time limitations and taking into account possible merging items that are not visible.
- Improved lane changes by enabling a greater deceleration limit in circumstances with a short deadline.
- Improved lateral control allows for more visibility when creeping forward.
- Improved modelling of road borders on roads with a significant degree of curvature to allow for more precise moves.
- Improved routing logic to eliminate needless detours/rerouting.
Advertisement
Interestingly, when quantifiable, Tesla provides actual data and percentage improvements in specific capacities.
Tesla increased static world detection by "up to 13%" by installing a new auto labelling system that utilised 165,000 fresh movies.
Tesla has relied significantly on a vast crew of thousands of "labelers" who literally watch movies and highlight what is seen in order to provide clean data for Tesla's machine learning systems.
However, the automaker has been creating technology to automatically label content for machine learning, and it appears as though it recently installed one that assisted in training this new FSD version 10.5.
Another feature that many people noticed was the inclusion of "emergency collision avoidance manoeuvres" in shadow mode.
Tesla appears to be expanding the vehicle's capability to conduct some more critical emergency moves to avert accidents, which were previously restricted due to Tesla's restriction on forceful steering manoeuvres in its driver-assist systems.
The function is now in "shadow mode," which means that the vehicles will not perform the manoeuvres themselves, but instead seek out opportunities to do so based on real-world events and transmit data to Tesla to determine how it would react.
This new Full Self-Driving Beta 10.5 software is scheduled to be evaluated for the next few weeks before Tesla releases a new version, which should allow more owners who paid for it to use the software.
The fleet's deployment has been much slower than CEO Elon Musk initially stated.
Advertisement