ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worst

@article{Bansal2019ChauffeurNetLT,
  title={ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worst},
  author={Mayank Bansal and Alex Krizhevsky and Abhijit S. Ogale},
  journal={ArXiv},
  year={2019},
  volume={abs/1812.03079}
}
Our goal is to train a policy for autonomous driving via imitation learning that is robust enough to drive a real vehicle. [] Key Method 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, and present ablation…

Figures and Tables from this paper

Exploring Imitation Learning for Autonomous Driving with Feedback Synthesizer and Differentiable Rasterization
TLDR
A novel feedback synthesizer is proposed for data augmentation that allows the agent to gain more driving experience in various previously unseen environments that are likely to encounter, thus improving overall performance and introducing task losses that penalize undesirable behaviors, such as collision, off-road, and so on.
Safer End-to-End Autonomous Driving via Conditional Imitation Learning and Command Augmentation
TLDR
The main contribution of this work is a recipe for building controllable imitation driving agents that improves upon multiple aspects of the current state of the art relating to robustness and interpretability.
Driver Dojo: A Benchmark for Generalizable Reinforcement Learning for Autonomous Driving
TLDR
This benchmark aims to encourage researchers to propose solutions that are able to successfully generalize across scenarios, a task in which current RL methods fail, and enables a better understanding of the impact of design decisions such as action and observation space on the generalizability of policies.
Symphony: Learning Realistic and Diverse Agents for Autonomous Driving Simulation
TLDR
S Symphony is proposed, which greatly improves realism by combining conventional policies with a parallel beam search, and factoring agent behaviour into goal generation and goal conditioning, and ensures that agent diversity neither disappears during adversarial training nor is pruned away by the beam search.
Iterative Imitation Policy Improvement for Interactive Autonomous Driving
We propose an imitation learning system for autonomous driving in urban traffic with interactions. We train a Behavioral Cloning (BC) policy to imitate driving behavior collected from the real urban
Causal Imitative Model for Autonomous Driving
TLDR
This paper proposes Causal Imitative Model (CIM), a model that explicitly discovers the causal model and utilizes it to train the policy, and shrinks the input dimension to only two, hence, can adapt to new environments in a few-shot setting.
Affordance-based Reinforcement Learning for Urban Driving
TLDR
This work proposes a deep reinforcement learning framework to learn optimal control policy using waypoints and low-dimensional visual representations, also known as affordances and demonstrates that its agents when trained from scratch learn the tasks of lane-following, driving around intersections as well as stopping in front of other actors or traffic lights even in the dense traffic setting.
SimNet: Learning Reactive Self-driving Simulations from Real-world Observations
TLDR
This work presents a simple end-to-end trainable machine learning system capable of realistically simulating driving experiences and is the first work that directly merges highly realistic data-driven simulations with a closed loop evaluation for self-driving vehicles.
Learning How to Avoiding Obstacles for End-to-End Driving with Conditional Imitation Learning
TLDR
A behavior-Cloning neural network architecture is proposed with the modified loss that enlarge the effects of errors for steer, which indicates the benefit to high an accuracy and the data augmentation of the image is crucial to the training of the proposed network.
Stabilization Approaches for Reinforcement Learning-Based End-to-End Autonomous Driving
TLDR
Two approaches to improve the stability of the policy model training with as few manual data as possible are proposed, which can converge quickly and stably in Gazebo in which previous methods can hardly converge.
...
...

References

SHOWING 1-10 OF 26 REFERENCES
Driving Policy Transfer via Modularity and Abstraction
TLDR
This work presents an approach to transferring driving policies from simulation to reality via modularity and abstraction, inspired by classic driving systems and aims to combine the benefits of modular architectures and end-to-end deep learning approaches.
DART: Noise Injection for Robust Imitation Learning
TLDR
A new algorithm is proposed, DART (Disturbances for Augmenting Robot Trajectories), that collects demonstrations with injected noise, and optimizes the noise level to approximate the error of the robot's trained policy during data collection.
Safe, Multi-Agent, Reinforcement Learning for Autonomous Driving
TLDR
This paper applies deep reinforcement learning to the problem of forming long term driving strategies and shows how policy gradient iterations can be used without Markovian assumptions, and decomposes the problem into a composition of a Policy for Desires and trajectory planning with hard constraints.
Deep Imitative Models for Flexible Inference, Planning, and Control
TLDR
This paper proposes Imitative Models, probabilistic predictive models of desirable behavior able to plan interpretable expert-like trajectories to achieve specified goals, and derives families of flexible goal objectives that can be used to successfully direct behavior.
Conditional Affordance Learning for Driving in Urban Environments
TLDR
This work proposes a direct perception approach which maps video input to intermediate representations suitable for autonomous navigation in complex urban environments given high-level directional inputs, and is the first to handle traffic lights and speed signs by using image-level labels only.
Imitating driver behavior with generative adversarial networks
TLDR
This paper adopts a method for overcoming the problem of cascading errors inherent in prior approaches, resulting in realistic behavior that is robust to trajectory perturbations, and extends Generative Adversarial Imitation Learning to the training of recurrent policies.
Virtual to Real Reinforcement Learning for Autonomous Driving
TLDR
A novel realistic translation network is proposed to make model trained in virtual environment be workable in real world, and is believed to be the first successful case of driving policy trained by reinforcement learning that can adapt to real world driving data.
DeepDriving: Learning Affordance for Direct Perception in Autonomous Driving
TLDR
This paper proposes to map an input image to a small number of key perception indicators that directly relate to the affordance of a road/traffic state for driving and argues that the direct perception representation provides the right level of abstraction.
Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car
TLDR
A method for determining which elements in the road image most influence PilotNet's steering decision is developed, and results show that PilotNet indeed learns to recognize relevant objects on the road.
Comparing human-centric and robot-centric sampling for robot deep learning from demonstrations
TLDR
It is observed in simulation that for linear SVMs, policies learned with RC outperformed those learned with HC but that using deep models this advantage disappears, and it is proved there exists a class of examples in which at the limit, HC is guaranteed to converge to an optimal policy while RC may fail to converge.
...
...