3 ms·

Report: Tesla using behavior cloning/supervised imitation learning for Autopilot

strangecosmos 8y ago
Key excerpt: “Tesla’s cars collect so much camera and other sensor data as they drive around, even when Autopilot isn’t turned on, that the Autopilot team can examine what traditional human driving looks like in various driving scenarios and mimic it, said the person familiar with the system. It uses this information as an additional factor to plan how a car will drive in specific situations—for example, how to steer a curve on a road or avoid an object. Such an approach has its limits, of course: behavior cloning, as the method is sometimes called… But Tesla’s engineers believe that by putting enough data from good human driving through a neural network, that network can learn how to directly predict the correct steering, braking and acceleration in most situations. “You don’t need anything else” to teach the system how to drive autonomously, said a person who has been involved with the team. They envision a future in which humans won’t need to write code to tell the car what to do when it encounters a particular scenario; it will know what to do on its own.” In this context, behavior cloning (or behavioral cloning) most likely means supervised imitation learning. That means training a neural network on a dataset of world states labelled with drivers’ actions. After training on these state/action pairs, the neural network attempts to predict the correct action for each state it encounters. (Detailed explanation: https://arxiv.org/abs/1701.06699 https://arxiv.org/abs/1701.06699) Here’s a great visualization of how Tesla currently sees world states: https://youtu.be/qgBBTDu1rzk https://youtu.be/qgBBTDu1rzk Waymo did a small-scale experiment with supervised imitation learning, which they wrote about here: https://link.medium.com/D5XORBIgQT https://link.medium.com/D5XORBIgQT They also published a paper: https://arxiv.org/abs/1812.03079 https://arxiv.org/abs/1812.03079 What do machine learning people think of this approach?

Key excerpt: “Tesla’s cars collect so much camera and other sensor data as they drive around, even when Autopilot isn’t turned on, that the Autopilot team can examine what traditional human driving looks like in various driving scenarios and mimic it, said the person familiar with the system. It uses this information as an additional factor to plan how a car will drive in specific situations—for example, how to steer a curve on a road or avoid an object. Such an approach has its limits, of course: behavior cloning, as the method is sometimes called… But Tesla’s engineers believe that by putting enough data from good human driving through a neural network, that network can learn how to directly predict the correct steering, braking and acceleration in most situations. “You don’t need anything else” to teach the system how to drive autonomously, said a person who has been involved with the team. They envision a future in which humans won’t need to write code to tell the car what to do when it encounters a particular scenario; it will know what to do on its own.” In this context, behavior cloning (or behavioral cloning) most likely means supervised imitation learning. That means training a neural network on a dataset of world states labelled with drivers’ actions. After training on these state/action pairs, the neural network attempts to predict the correct action for each state it encounters. (Detailed explanation: https://arxiv.org/abs/1701.06699 https://arxiv.org/abs/1701.06699) Here’s a great visualization of how Tesla currently sees world states: https://youtu.be/qgBBTDu1rzk https://youtu.be/qgBBTDu1rzk Waymo did a small-scale experiment with supervised imitation learning, which they wrote about here: https://link.medium.com/D5XORBIgQT https://link.medium.com/D5XORBIgQT They also published a paper: https://arxiv.org/abs/1812.03079 https://arxiv.org/abs/1812.03079 What do machine learning people think of this approach?