End-to-End Driving Via Conditional Imitation Learning
By Felipe Codevilla, Matthias Müller, Alexey Dosovitskiy, Antonio M. López, Vladlen Koltun. 2017
1,304 citations recorded by the source corpus.
Deep networks trained on demonstrations of human driving have learned to follow roads and avoid obstacles. However, driving policies trained via imitation learning cannot be controlled at test time. A vehicle trained end-to-end to imitate an expert cannot be guided to take a specific turn at an upcoming intersection. This limits the utility of such systems. We propose to condition imitation learning on high-level command input. At test time, the learned driving policy functions as a chauffeur that handles sensorimotor coordination but continues to respond to navigational commands. We evaluate different architectures for conditional imitation learning in vision-based driving. We conduct experiments in realistic three-dimensional simulations of urban driving and on a 1/5 scale robotic truck that is trained to drive in a residential area. Both systems drive based on visual input yet remain responsive to high-level navigational commands.
Subfields: Artificial Intelligence, Automotive Engineering, Control and Systems Engineering, Reinforcement Learning, Robotics & Control
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Sergey Levine
- Pieter Abbeel
- Chelsea Finn
- Yoshua Bengio
- Mohit Bansal
- Trevor Darrell
- Ruslan Salakhutdinov
- Dacheng Tao
In Automotive Engineering
- Andreas Geiger
- Sergey Levine
- Kashyap Chitta
- Pravendra Singh
- Pranav Singh Chib
- Raquel Urtasun
- Hongyang Li
- Li Chen
In Control and Systems Engineering
- Sergey Levine
- Pieter Abbeel
- Chelsea Finn
- Silvio Savarese
- Li Fei-Fei
- Abhishek Gupta
- Dieter Fox
- Vladlen Koltun
In Reinforcement Learning
- Sergey Levine
- Pieter Abbeel
- Chelsea Finn
- Aviral Kumar
- Abhishek Gupta
- Benjamin Eysenbach
- Shixiang Shane Gu
- Justin Fu
In Robotics & Control
- Sergey Levine
- Pieter Abbeel
- Chelsea Finn
- Silvio Savarese
- Vladlen Koltun
- Li Fei-Fei
- Dieter Fox
- Andreas Geiger
Citations in the corpus
- Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems (2020, background): “Imitation learning has been a popular approach towards end-toend, data-driven methods in autonomous driving (Bojarski et al., 2016; Sun et al., 2018a; Pan et al., 2017; Codevilla et al., 2018).”
- A Survey of Deep Network Solutions for Learning Control in Robotics: From Reinforcement to Imitation (2016, neutral): “Codevilla et al. (2017) observes that the pure behavior cloning assumption could break under certain situations, such as when a driver approaches an intersection.”
- Driving Policy Transfer via Modularity and Abstraction (2018, background): “A number of works employ specialized learning techniques and network architectures to facilitate transfer: for example, variants of domain adaptation [43, 14, 46, 3, 31] and specialized architectures [36, 47].”
- Learning to Generalize Across Long-Horizon Tasks from Human Demonstrations (2020, background): “, instructions [8], video demonstrations [41, 16, 10], or goal observations [6, 25].”
- BADGR: An Autonomous Self-Supervised Learning-Based Navigation System (2020, background): “Methods based on imitation learning have been demonstrated on real-world robots [16], [17], but again depend on humans for expert demonstrations, which does not constitute a continuously self-improving system.”
- Composable Action-Conditioned Predictors: Flexible Off-Policy Learning for Robot Navigation (2018, background): “Prior work has aimed to achieve similar results by conditioning the policy on a goal or objective indicator [4, 5].”
- Deep Imitative Models for Flexible Inference, Planning, and Control (2018, methodology): “While several works augmented IL with goal conditioning (Dosovitskiy & Koltun, 2016; Codevilla et al., 2018), these goals must be specified in advance during training, and are typically simple (e.”
- Simitate: A Hybrid Imitation Learning Benchmark (2019, neutral): “focusing on imitation learning target autonomous driving [33], [34].”
- CARLA: An Open Urban Driving Simulator (2017, methodology): “Table S.1 details the configuration of the network used in the imitation learning approach [4].”
- Gradient Surgery for Multi-Task Learning (2020, background): “While we studied multi-task supervised learning and multi-task reinforcement learning in this work, we suspect the problem of conflicting gradients to be prevalent in a range of other settings and applications, such as meta-learning, continual learning, multi-goal imitation learning [10], and multi-”
- Continual Learning for Robotics (2019, background): “This as well happens for autonomous robots that learn new manipulation tasks, for exploration policies, as well as for autonomous vehicles that need to learn and adapt to new circumstances (Bojarski et al., 2016; Codevilla et al., 2017; Jaritz et al., 2018; Rhinehart et al., 2018).”
- Language-Conditioned Imitation Learning for Robot Manipulation Tasks (2020, background): “Beyond robotics, the combination of vision and language has received ample attention in visual question-and-answering systems (VQA) [22, 5] and vision-and-language navigation (VNL) [34, 18, 10].”
- Reward-Conditioned Policies (2019, background): “In contrast, supervised learning is comparatively well understood, and even imitation learning methods can often provide a much simpler approach to learning effective policies when demonstration data is available [3, 27, 33].”
- Exploring the Limitations of Behavior Cloning for Autonomous Driving (2019, methodology): “end-to-end urban driving based on CIL [12].”
- MP3: A Unified Model to Map, Perceive, Predict and Plan (2021, background): “First, we see that only by using the discrete high-level action (Action only), which is the same information as the CIL and CNMP baselines use, our model is able to achieve a much better route following than those.”
- ST-P3: End-to-end Vision-based Autonomous Driving via Spatial-Temporal Feature Learning (2022, background): “One direction of end-to-end pipelines is to focus on the ultimate planning task mainly without explicit design of the intermediate representation [13, 14, 42, 9, 12, 26].”
- Conditional Vehicle Trajectories Prediction in CARLA Urban Environment (2019, methodology): “Table 2: Infraction distance comparison (in km between infractions, more is better), with baselines [8, 30].”
- Planning-oriented Autonomous Driving (2022, background): “Pioneering works [14,16,21,22,78,95, 97,106] verified this vanilla design in the closed-loop simulation [26].”
- Neural Autonomous Navigation with Riemannian Motion Policy (2019, background): “An alternative scheme is to train an end-to-end network [16], [17], [18] that directly outputs the control commands.”