AI Predicts Other Drivers’ Behavior

SUGGESTED NEWS

It could help self-driving cars predicts what other cars Drivers, pedestrians, and bikes will do in the future. People are one of the main problems with cars that can drive themselves in cities. To drive in Boston, a robot must be able to predict how other cars, people, and bikes will move. To avoid people, the robot just parks the car. However, current AI systems can only predict the next move of one agent (roads typically carry many users at once.)

Researchers at MIT have found an easy answer. A problem with predicting how multiple agents will act can be solved quickly by breaking it up into smaller parts.

Simulator of AI driving

Using AI to copy driving on the road.
The researcher can Predicts Other Drivers with blue cars will move in the future by simulating how they will move around other cars, bikes, and people. MIT

To do this, they must first guess which car, cyclist, or pedestrian has the right of way and which agent will give way.

Waymo said that these predicted routes were more accurate than other methods of machine learning. A recent Waymo model was beaten by the method from MIT. They could remember less because they split up the work.

“This is a simple idea that no one has fully looked into.” It’s a plus that it’s easy to understand. Co-author Xin “Cyrus” Huang, is a graduate student in the Department of Aeronautics and Astronautics and a research assistant in Brian Williams’ lab, which is part of the CSAI Laboratory.

The article was written by Hang Zhao, an assistant professor at Tsinghua University, Qiao Sun, a research assistant, Junru Gu, a graduate student, and Junru Gu, also a graduate student. Computer Vision and Pattern Recognition Conference, or CvPR Conference

The little

In a traffic situation like a four-way intersection, you need a map and a record of how cars, bikes, and people have interacted with each other in the past.

A relation predictor looks at these pieces of information to figure out which of the two agents will pass first. Since the passing agent is independent, a marginal predictor can figure out where it will go.

A conditional predictor predicts that an agent will give in. The system plans for multiple yielders and passer courses figure out how likely they are to happen, and picks the six that are most likely to happen.

This is how M2I figures out what will happen in the next eight seconds. So, a car would slow down to let a person cross, and then speed up again when the person was out of the way. Another time, a car went from a small street to a major road.

By combining marginal and conditional predictors, M2I might be able to Predicts Other Drivers’ behavior

Real tests of driving

The lidar sensors and cameras on Waymo’s self-driving cars took millions of photos that are in the Open Motion Dataset. They were experts in cases with more than one person.

To see how well the methods worked, scientists compared the actual paths of cars, bikes, and people to the paths of the six predicted samples. It also did better than the baseline models in the “overlap rate,” which is a way to measure how often two objects bump into each other. Then, M2I.

We tried to solve this problem by acting more like people. No one can know what will happen. Huang says, “We move quickly.”

M2I makes the problem easier to understand so that the user can understand what the model is doing. Huang says that people will trust self-driving cars in the long run.

Two cars move forward at a four-way stop, but the drivers don’t know who should give way.

On the list of things to do. In addition to making fake driving data, their method could also help improve how well models work.

“It’s hard to predict what will happen in the future when many agents interact. Professor Masayoshi Tomizuka and Wei Zhan, both of UC Berkeley’s Mechanical Engineering Department, say that M2I’s relation predictor can tell the difference between agents predicted marginally or conditionally. The prediction model could show agent interactions. Exclusion of two companions.

Originally published at https://scitechdaily.com/news/technology/

Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Learn how your comment data is processed.

0 Comments
Inline Feedbacks
View all comments

Recent Articles