RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2

Tesla FSDC (Full Self-Driving Computer)

Product
Developers: Tesla Motors, Samsung Electronics
Date of the premiere of the system: 2019/04/22
Branches: Transport,  Electrical and Microelectronics
Technology: Processors

Content

Main article: Autopilot (self-driving car)


Tesla FSDC (Full Self-Driving Computer) is a processor called the "fully autonomous computer" in the company. It is a high-performance specialized chip developed by the Samsung laboratory in Texas by order of Tesla exclusively to ensure autonomous safe movement of the car.

2023

Number of fatal crashes involving Tesla Autopilot rises 6-fold

In early June 2023, it became known the sharply increased number of accidents involving cars Tesla with the Tesla Autopilot driver assistance system. The number of deaths and serious injuries also increased significantly. In June 2022, only three fatal autopilot accidents were reported, and a year later the number of such cases increased to 17. The total number of accidents involving Tesla cars when using the Autopilot system has reached 736 accidents since 2019, according to the National Highway Traffic Safety Administration. USA

US Secretary of Transportation Pete Buttigieg in May 2023 said in an interview with the Associated Press that, in his opinion, the name "Autopilot" misleads users. Technology from Tesla is not self-driving technology, but new figures suggest that this is how people treat it, which sometimes leads to deadly tragedies.

Number of fatal crashes involving Tesla autopilot rises 6-fold in a year

Washington Post journalists interviewed auto experts who were unanimous in their conclusions. The rise in autopilot-related crashes could be a result of changes made by Tesla CEO Elon Musk, including the removal of radar sensors from new Model 3 and Model Y vehicles in 2021, and from Model S and Model X in 2022, they said.

The U.S. National Highway Traffic Safety Administration (NHTSA) reminds the public that all advanced driver assistance systems require a human driver to be in control at all times and fully engaged in the driving process. Accordingly, the laws of all states of America make the human driver responsible for driving vehicles for June 2023.

NHTSA collects detailed crash data using driver assistance technology from 2021. Nearly all of the 807 automation-related crashes in this dataset were involving a Tesla car. Subaru came in second with 23 cases. NHTSA highlighted that four of the 17 Tesla-related fatalities involved a motorcycle and one in an ambulance accident. [1]

Recall of 360,000 defective electric vehicles due to faulty autopilot

On February 16, 2023, Tesla announced the recall of more than 360,000 of its electric vehicles due to faulty software. The US National Highway Traffic Safety Administration (NHTSA) explained that we are talking about software that allows an electric car to exceed the maximum permitted speed and pass intersections in an illegal or unpredictable way. The department added that the Full Self-Driving (FSD) Beta system may not respond satisfactorily to changes in the established speed limits or not fully take into account speeding by the driver. Read more here.

2022

Autopilot sees no child pedestrians

In early August 2022, researchers from The Dawn Project conducted a test to determine by Tesla autopilot various stationary objects. In a video published on a YouTube channel, an electronics-driven car does not notice the child's dummy and knocks him down several times.

The test was attended by an electric car Tesla Model 3 equipped with the latest version of ON Tesla Full Self-Driving (FSD). The researchers added that there was good sunny weather outside without precipitation. The experiment itself was conducted under controlled conditions at a test site in. California And its results were confirmed by the National Highway Safety Administration, USA which is already conducting an open and active investigation of the system from autopilot Tesla on August 11, 2022.

"It's a deadly threat." Tesla autopilot 'doesn't see' child pedestrians

In several tests, a professional test driver discovered that the software, released in June 2022, failed to recognize the child's figure at an average speed of 40 km per hour, after which the car was hit by a mannequin. The Dawn Project founder Dan O'Dowd called the results deeply disturbing. More than 100,000 Tesla drivers are already using full self-driving on public roads, putting children at huge risk in communities across the country. O'Dowd argues that the test results show the need to ban self-propelled vehicles until the company proves those vehicles will not hit children at crosswalks.

Tesla Full Self-Driving

According to the FSD developers, the autopilot works using ultrasonic sensor systems, a set of radars and all-round cameras with a different sector and depth of video capture. Various volumetric scanning sensors control the area around the car 360 ° at a distance of up to 250 m and allow you to detect objects of any shape and size, both stationary and moving. However, despite these assurances, the autopilot did not notice the child's stationary mannequin on the road and hit him three times, continuing to move on[2]

Tesla electric car got into an accident again due to autopilot failure

In early February 2022 electric vehicle Tesla , he got into a traffic accident due to a failure autopilot using the Full Self-Driving Beta program. The car that collided with the pole was caught on video. More. here

2021: Tesla's new autopilot slows down sharply due to non-existent obstacles

In mid-November 2021, Tesla notes an increase in complaints of serious and dangerous cases of phantom braking in the autopilot system in the latest software updates. The system falsely detects objects on the road or calculates false collision trajectories that will not actually be, and then emergency braking of the car is activated to try to avoid a phantom obstacle.

Phantom braking, is a term used to describe a situation where an advanced driver assistance system or self-propelled system hits the brakes for no good reason. Braking without warning, this is what the user wants to avoid, as it can lead to a more severe accident already for other motorists. For Tesla owners, this has always been part of the autopilot system, but for the most part it was manageable and the cases themselves rarely happened, but since the beginning of September 2021 for many Tesla owners, the situation has deteriorated greatly.

Tesla's new autopilot slows down sharply due to non-existent obstacles

In October 2021, Tesla briefly stopped the new version of its full self-driving (FSD) software after many testers reported constant problems with phantom braking, CEO Elon Musk admitted this problem. FSD is being tested by only a limited number of car owners, but it has now become clear that phantom braking is becoming a major problem for autopilot users as well. Since the beginning of November 2021, this has been confirmed by the US National Highway Traffic Safety Administration ( NHTSA), the ministry notes a significant increase in the number of complaints from Tesla owners about the phantom braking problem.

File:Aquote1.png
After accepting the delivery at the end of May 2021, we gained 14 thousand km by car and got a terrible experience with cruise control, taking into account the road situation, which for no apparent reason presses the brake when there is nothing or passing cars ahead. The behaviour can be smooth at 8-16km/h or in some cases full brake pressure at 60-90km/h, putting us in danger of a collision from behind. Several times we were already close to this, but other drivers reacted quickly, "NHTSA said in a recent complaint.
File:Aquote2.png

All complaints are very similar to this: drivers say their cars experience a significant amount of phantom braking in autopilot. Some of the owners reported that they had contacted Tesla about the issue, but were told it was due to a software update. As noted by Electrek, the increase in the number of complaints began around the period when Tesla switched to an autopilot system based on machine vision in update 2021.40 and abandoned the use of radar.[3]

2019: Details about Tesla FSDC self-driving car processor

Tesla presented details about its own processor for use in autonomous cars for the operation of the software part of the platform, which became known on April 23, 2019. The total performance of the onboard system on FSDC chips in the company is estimated at 144 TOPS (trillions of operations per second, teraOps per second).

Processor: Tesla Full Self-Driving Computer. Illustration: Tesla

The developers of the chip took care of its reliability, ensuring it, first of all, due to redundancy: FSDC consists of two completely identical systems located on a single board. Both systems have their own power supplies and function independently of each other. In the process of movement, both systems process, data GPS radar, maps, rudder position, data from other sensors and sensors. Then both systems process the information and compare and confirm the relevance of both results. In practice, in case of damage or failure of one of the systems, a special test will ON detect problems and isolate it, while common power circuits will not storage systems be affected.

FSDC chips are available in a 2116-pin BGA case (array of ball pads for soldering) with dimensions of 37.5 x 37.5 mm. The internal organization of the FSDC crystal has 12,464 C4 contact pads, the crystal itself is made using 12-level metallization.

The chip is manufactured in compliance with the 14 nm CMOS FinFET process standards and contains 250 million logic gates based on 6 billion transistors. The area of ​ ​ the FSDC crystal is 260 square meters. mm

The chip is capable of processing video data from cameras with a flow of up to 2.5 billion pixels per second. The internal 24-bit image processor bus, responsible, among other things, for improved markup processing and noise reduction, has a performance of up to 1 billion pixels per second.

The chip directly integrates a high-speed interface for working with LPDDR 4 memory via a 128-bit bus with a bandwidth of 4266 Gbps and peak performance of up to 68 GB/s.

Almost half of the area of ​ ​ the FSDC crystal is occupied by two neuroprocessors. The performance of each in Tesla is estimated at 36 TOPS at a clock speed of 2 GHz, which in total allows us to talk about the performance of up to 72 TOPS per chip, which significantly exceeds the company's initial task of achieving the performance of a neuroprocessor of 50 trillion operations per second.

For general computing, 12 64-bit ARM A72 cores operating at 2.2 GHz are integrated into each FSDC chip.

The chip also features a built-in hardware H265 codec and a set of hardware validation modules that track input and output.

A separate integrated cryptographic module digitally signed by Tesla prevents attempts to hack the vehicle's on-board system, filters any attempts to interfere or enter suspicious visual information, for example, to deceive the car about the presence of a pedestrian, or reconfigure the output to bypass security measures.

The performance of the neuroacselector at Tesla was called sufficient for processing data of all sensors of the system - due to "special mathematical operations" in FSDC (mainly addition and multiplication operations), which provide "huge advantages in performance and economy," since this data does not need to be stored in DRAM - for this, the neuroprocessor is equipped with 32 MB of faster SRAM.

According to Tesla's official presentation, the power consumption of the entire FSDC system does not exceed 100 watts, or about 50 watts per computing module.[4]

Notes