MIT is trying to develop an AI model that can drive almost like a human
Random Tech News

MIT is trying to develop an AI model that can drive almost like a human


Creating driverless cars capable of reasoning and handling situations like humans is one of the most important long-term goals of companies pioneering autonomous vehicles. such as Waymo, GM, Cruise, Uber and some other big names. Mobileye, a prominent enterprise in vision technology development for Advanced Driver Assistance Systems (ADAS) and autonomous driving, under Intel Corporation, has previously proposed a mathematical model with Responsibility-Sensitive Safety (RSS) name. This invention is described as an advanced way for the car to make more precise decisions during operation, much like the way we normally re-ride, and has received quite a few hits. positive prices from the community side. In addition, another big man who is also very interested in the autonomous vehicle sector is that Nvidia is also actively developing Safety Force Field – a “policy” to help the car make accurate decisions based on on the ability to track and evaluate any unsafe actions during operation through real-time analysis of sensor data.

  • Bosch introduces the sensor box designed for unmanned “flying taxis”

The above is a typical example of the commendable efforts of businesses to focus on developing safe driving capabilities – the key factor determining the success of an autonomous vehicle. However, scientists at the Massachusetts Institute of Technology (MIT) have a much bigger ambition, which is to build an AI system capable of driving exactly like a human. More specifically, a team of MIT artificial intelligence experts are currently working on a way to leverage map data and other types of visual data like GPS to enable AI-based autonomous vehicles to have. can learn how to drive a human vehicle, and at the same time “fluently” apply trained knowledge to intricately planned, test routes in environments never seen before. Accordingly, this work is built on the foundation of end-to-end navigation systems designed by Daniel Rus, director of the American Laboratory of Computer Science and Artificial Intelligence (CSAIL). , and is expected to be presented at the International Conference on Robotics and Automation in Long Beach, California, USA next month.

Transportation in the future

  • Safety features in Tesla cars that make you “want” to go the wrong lane are also difficult

The previous model of machine learning by Daniel Rus and his colleagues was rated as “cumbersome” and less likely to be deployed extensively. However, in this new model, MIT scientists have almost completely overcome the above problem. “With our system, you don’t have to train the car on all roads, but you can download a new map for the vehicle to help the vehicle navigate through the roads it has. had never come across before, said engineer Alexander Amini, who heads the research team.

As engineers Amini and colleagues explained, their AI systems are capable of observing and learning to control a vehicle from the driver, and then correlating the steering wheel’s revs with each curve. It can be viewed through the camera system and maps that have been introduced into the system earlier. Finally, the car can give exact “driving orders” for each different road, such as a straight line, a four-way intersection or a junction, rotation …

AI is capable of driving

  • 55 self-propelled cars from 5 major car manufacturers are being tested at the same time

In many experiments, the researchers provided the machine learning model with a map with a randomly selected route. During vehicle operation, the system has extracted visual features from the camera, allowing it to predict each element of the structure of the road such as signs or speed bumps, hard separators … Read More Furthermore, the AI ​​model is also capable of correlating visual data with map data to indicate inadequacies, helping to more accurately identify the vehicle’s position on the road and ensure the vehicle is on the route. safest way. For example, when the AI ​​model was driving the car on a straight road and there were absolutely no turns, but the map system went wrong and indicated that the car needed to turn right, the AI ​​recognized This is unreasonable and knows how to control the vehicle on the correct track without following orders from the map.

Self-propelled vehicle

  • Are humanity on the right track in developing autonomous vehicle technology?

“In the real world, sensors can’t always achieve perfect accuracy. However, we want to ensure that our model can “adapt” to the various errors that arise in the actual operation, by building a system that can recognize anomalies and still capable of navigating as well as accurately locating on all roads ”, added engineer Alexander Amini.

.

Leave a Reply

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