(T) I had the opportunity to attend at the ODSC West conference, a talk from Waymo’s engineering manager’s Chen Wu. Following is a summary of what I learned about the talk, and ChauffeurNet. First, Waymo has a long list of models for its self-driving cars for perception, prediction, planning, and simulation. Second regarding the data sources, I was surprised to learn that a lot of the training data comes from manual labels either processed internally and outsourced. Waymo believes that manual labeling outperforms self-supervised learning, weakly supervised learning, or semi-supervised learning for large number of labels! Third regarding the models, the challenge of self-driving cars is the very large number of models and the time to find the right architecture for a deep learning model:
- “Conducting an end-to-end search ordinarily requires exploring thousands of architectures manually, which carries large computational costs. Exploring a single architecture requires several days of training on a data center computer with multiple GPU cards, meaning it would take thousands of days of computation to search for a single task!”
However, the solution that Waymo implemented to that challenge was AutoML with reinforcement learning and transfer learning:
- Neural Architecture Search with Reinforcement Learning
- Learning Transferable Architectures for Scalable Image Recognition
Waymo has demonstrated a very interesting prototype of a model, called ChauffeurNet, to learn path for the car through imitation planning:
- “Our goal is to train a policy for autonomous driving via imitation learning that is robust enough to drive a real vehicle…
We find that standard behavior cloning is insufficient for handling complex driving scenarios…
We propose exposing the learner to synthesized data in the form of perturbations to the expert’s driving, which creates interesting situations such as collisions and/or going off the road. Rather than purely imitating all data, we augment the imitation loss with additional losses that penalize undesirable events and encourage progress – the perturbations then provide an important signal for these losses and lead to robustness of the learned model…
We show that the ChauffeurNet model can handle complex situations in simulation.”
- ChauffeurNet’s Inputs and outputs:
- ChauffeurNet’s models:
- “We presented our experience with what it took to get imitation learning to perform well in real-world driving. We found that key to its success is synthesizing extreme situations around the expert’s behavior and mixing real and simulated data that discourage undesirable (driving) car behaviors…That said, the model is not yet fully competitive with (exiting Waymo?) motion planning approaches (that use a combination of machine learning and hand-coded rules) but we feel that this is a good step forward for machine learned driving models…We believe that augmenting the expert demonstrations perhaps within a reinforcement learning framework, will be the key to improving the performance of these models.”
A video from ChauffeurNet presented at Google I/0:
Note: The picture above is from Waymo.
Copyright © 2005-2019 by Serge-Paul Carrasco. All rights reserved.
Contact Us: asvinsider at gmail dot com.
Categories: Artificial Intelligence, Autonomous Vehicles