A complete autonomous driving process includes three levels of "perception-decision-execution", in the perception layer, with sensors configured as standards, the industry is divided into two camps: the vision system and the radar system. Tesla belongs to the camera + millimeter wave radar as the main sensor, the essence is the "light data heavy algorithm" of the visual system. The radar system adopts multi-sensor fusion solutions such as lidar, camera, and millimeter-wave radar, which is essentially "heavy data and light algorithm".
tesla
Today we are mainly going to talk about Tesla's Full Self-Driving (FSD, fully autonomous driving system).
Full Self-Driving (FSD) is an advanced driver assistance suite on Tesla vehicles that includes autopilot and some other more complex functions: 1. Active Cruise Control; 2. Automatic Assist steering; 3. AutomaticAlly Assisted Lane Change; 4. Auto-Assisted Navigation; 5. Automatic Parking; 6. Summon; 7. Intelligent Summon; 8. Active Safety Feature; 9. Display and Respond to Traffic Lights and Stop Signs; 10. Automatic assisted driving on urban streets; 11. Fully autonomous driving. Its basic principle is to accurately identify various objects and objects on the road, and its driving computer will make various driving control actions according to these identified road conditions.
One concept corresponding to the FSD is the Tesla "Autopilot," which is also the standard suite of driver assistance for Tesla's new vehicles, with features including automatic emergency braking, forward collision warning, lane keeping, and adaptive cruise control.
In contrast, although the standard "Autopilot" is called "automatic assistance system", the key point is "assist", while the key point of FSD is "fully automatic driving". According to the grading standard of autonomous driving, Autopilot is equivalent to L2 level, and FSD is equivalent to L4 level. But in fact, FSD is still some distance away from "fully autonomous driving".
Key improvements to the latest FSD Beta 10.6 include:
Improved object detection network architecture for non-VRU (e.g., cars, trucks, buses). A 7% increase in recalls, a 16% reduction in depth errors in crossover inspections and a 21% reduction in speed detection errors.
- The average relative error of the new visibility network is reduced by 18.5%.
- At high curvature and at night, the detection accuracy of the new universal static target network has been improved by 17%.
- Using object prediction capabilities other than intersections, the parking position for unprotected left turns is improved, while better responding to oncoming targets.
- By integrating modeling of the ends of the parallel area, more space is provided for vertical alignment during parallel.
- Improved comfort when offsetting targets that are cutting out of the lane.
Meanwhile, Tesla has repeatedly said that the speed of improvements to its supercomputer, the Dojo, will increase.
However, after the launch, it was also a frequent problem in user testing, encountering many unexpected situations such as map delays, automatic driving into public transportation lanes, and almost hitting pedestrians when turning. There is still a long way to go before it can reach true fully autonomous driving. But even so, Tesla's self-driving technology is still attractive to consumers.
Can Tesla be the car of the future in your heart? How many more five years will it take Musk's pie to become a reality?