Indian American Researchers Help Erase “Blind Spots” In Self-Driving Cars

Indian American researchers have created a model to detect artificial intelligence “blind spots” in self-driving cars through human inputs. It is supposed to reduce some risky errors in these vehicles in reality.

Coming from Microsoft and MIT, the researchers designed the model that can realize and analyze the cases when there is any difference between the training examples and the real-life application of autonomous systems. With this model, engineers can enhance some factors concerning AI systems’ safety, including autonomous robots and self-driving cars.

The AI systems which are built into self-driving cars are conducted comprehensively in virtual simulations to ensure that these cars can face almost all of the possible circumstances on the road. However, a few sudden errors may occur beyond developers’ expectations.

The researchers checked the model by a series of video games and a simulated person, which can adjust the path of the character on the screen.

In the second step, they combine the model with conventional testing and training methods designed for unmanned robots and cars, along with feedback from humans.

In addition to Ramya Ramakrishnan, the two other authors of this research include Julie Shah, Ece Kamar, Eric Horvitz, and Debadeepta Dey. Among those, Shah works as both the Interactive Robotics Group head of the CSAIL. They all are researchers of Microsoft Research.