Saturday , October 23 2021

Lyft collects driver data to train its autonomous vehicle systems

Lyft this morning announced It has started using data from its hail network to improve the performance of its autonomous vehicle systems. A subset of driver cars – currently Select Express Drive vehicles as well as Lyft's autonomous vehicles in Palo Alto and selected vehicles that follow the vehicles for safety reasons – are now equipped with low-cost camera sensors with which they can capture challenging scenarios while helping to solve problems such as generating 3D maps and improving simulation tests.

Lyft, one of the companies forced to abandon driverless vehicle testing because of the pandemic, is trying to support development as much of its fleet and Palo Alto pilot remain on the ground. While VentureBeat said in an earlier interview that it would "double" simulation using data from the approximately 100,000 miles traveled by its self-driving cars, there is a limit to the possibilities of simulation.

Partly as a result of halted real vehicle tests, the corona virus has delayed Lyft rival Waymo's work by at least two months. In the meantime, Ford has driven the start of its driverless vehicle service from 2021 to 2022. Analysts like Brian Collie of the Boston Consulting Group now believe that autonomous cars will not be fully commercialized until at least 2025 or 2026 3 years later than originally assumed.

According to Lyft, driver vehicle data can continuously update the company's 3D maps that the company has created using technology developed by Blue Vision LabsLike other outfits that develop self-driving vehicle systems, Lyft creates high-resolution maps of streets, buildings, vegetation and other objects at a centimeter height to make it easier to locate vehicles. These maps also contain contextual information such as speed limits and the position of lanes and pedestrian crossings. The Lyft backend generates this context information from ridesharing opportunities by using a combination of computer vision and AI to automatically identify objects of traffic (e.g. traffic lights). This is combined with situation data, e.g. B. where lanes and traffic lights are located to better understand how drivers deal with risk situations.

According to Lyft, every driver in the program receives one-sided disclosure with information about the camera and the data collected. The camera is in no way connected to the driver – it faces forward and does not collect audio.

Data from the Lyft network – along with visual localization technology – also help illuminate human driving patterns, the company said. Lyft tracks the trajectories of real drivers on its maps with “great accuracy” to ensure, for example, that its autonomous vehicles maintain optimal lane positions. “Thanks to carpooling, our motion planner (autonomous vehicle) doesn't have to use ad hoc heuristics like tracking lane centers to decide where to go. This requires various exceptions to handle all possible curve cases, ”the company said in a blog post.

"Instead, the planner can rely on the real information and … the human driving experience, which of course is included in the ridesharing options," continues the article. “While common sense indicates that it is safest to stay near the center of the lane, historical data (carpooling) shows that this assumption is not always true. Human driving is much more nuanced due to local features on the road (such as parked cars or potholes) and other aspects such as road design or shape and view. "

This approach prompted Lyft to adopt an autonomous system design paradigm called "man-inspired" planning, initially in a Press release last december. Lyft's planning system uses carpooling to learn how cars that perform high-speed merges can slow down, and checks the safety and legality of planned behaviors before they are carried out. This is similar to Nvidia's Safety Force Field and the responsibility of the Intel subsidiary Mobileye. Sensitive security. The concept of perceived safety is also taken into account, which refers to minimizing the perception of passengers and other drivers (e.g. increasing the distance to a guide car or ensuring that the autonomous car does not come too close to a lane divider). Passenger comfort is also taken into account – for example the reduction in speed in bends that can cause nausea – and the efficiency of the route.

Lyft self-driving

“Every day, trips are carried out in our network that cover a wide range of driving scenarios, from pick-up and delivery services to situations that require immediate and critical thinking. However, since autonomous vehicles (AVs) are becoming a common transportation option, performing such real-time assessments is no longer limited to human drivers, ”wrote Lyft. "By using (ride-hail) data, Lyft is in a unique position to develop safe, efficient and intuitive self-driving systems."

In a way, Lyft's approach is similar to Tesla's driverless vehicle testing using simulations, test tracks, and public roads, but also "shadow-tested" the capabilities of its cars by collecting billions of miles of data from hundreds of thousands of data from customer-owned vehicles "during normal driving ". Tesla's autopilot – the software layer that runs on the custom chips – is an advanced driver assistance system (ADAS) that uses machine learning algorithms and a range of cameras, ultrasound sensors, and radars to perform self-parking, lane-centering, and adaptive cruises. Autobahn changes and other services. The company had previously claimed that cars with full self-driving capabilities, a premium autopilot package, would one day be ready for "automatic driving on city streets" and for "recognizing and responding to traffic lights and stop signs".

Lyft self-driving

The research and development department behind Lyft's efforts – Level 5 – was founded in July 2017 and developed, among other things, novel 3D segmentation frameworks, methods for evaluating energy efficiency in vehicles and techniques for tracking vehicle movement using crowdsourcing cards. Last year, Lyft announced the opening of a new testing facility in Palo Alto, California, near the headquarters of the Level 5 division. This development occurred after a year in which Lyft expanded access to its self-drive service for employees in Palo Alto with drivers for human security in a limited area.

In November 2019, Lyft announced that its autonomous cars traveled four times more quarterly than six months earlier and that around 400 employees worldwide are responsible for development (out of 300). In May, the company partnered with Waymo, Google’s parent company, to enable customers to access Waymo driverless cars from the Lyft app in Phoenix. And Lyft works continuously with the self-driving autostart Aptiv, which provides Lyft customers in Las Vegas with a small fleet of autonomous vehicles.

About Nancie Clifford

Nancie Clifford is a housewife and loves technology. He writes on various websites.

Check Also

Warehouse workers sue Amazon for COVID-19 exposure after the death of an employee's relative

An Amazon Prime billboard in New York City. (GeekWire Photo / Taylor Soper) Barbara Chandler …

Leave a Reply

Your email address will not be published. Required fields are marked *