Exploring Data Aggregation in Policy Learning for Vision-Based Urban Autonomous Driving

Prakash, Amit; Behl, Aseem; Ohn-Bar, Eshed; Chitta, Kashyap; Geiger, Andreas

doi:10.1109/cvpr42600.2020.01178

Cited by 62 publications

(53 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After that, CIL [10] and CILRS [11] addressed directional multimodality in AD by using branched action heads where the branch is selected by a high-level directional command. While the aforementioned methods are trained via behavior cloning, DA-RB [36] applied DAGGER [40] with critical state sampling to CILRS. Most recently, LSD [32] increased the model capacity of CILRS by learning a mixture of experts and refining the mixture coefficients using evolutionary optimization.…”

Section: Related Workmentioning

confidence: 99%

“…Our training scheme using Roach can be applied to improve the performance of existing IL agents. Here we use DA-RB [36] (CILRS [11] + DAGGER [40]) as an example to demonstrate its effectiveness.…”

Section: Il Agents Supervised By Roachmentioning

confidence: 99%

“…Implementation Details: Our implementation follows DA-RB [36]. We choose a Resnet-34 pretrained on ImageNet as the image encoder to generate a 1000-dimensional feature given i RL ∈ [0, 1] 900×256×3 , a wide-angle camera image with a 100 • horizontal FOV.…”

Section: Cilrsmentioning

confidence: 99%

“…Even though nowadays, most autonomous driving (AD) stacks [30,48] use individual modules for perception, planning and control, end-to-end approaches have been proposed since the 80's [35] and the success of deep learning brought them back into the research spotlight [5,50]. Numerous works have studied different network architectures for this task [3,16,52], yet most of these approaches use supervised learning with expert demonstrations, which is known to suffer from covariate shift [36,40]. While data augmentation based on view synthesis [2,5,35] can partially alleviate this issue, in this paper, we tackle the problem from the perspective of expert demonstrations.…”

Section: Introductionmentioning

confidence: 99%

“…As a real-world example, Pan et al [34] demonstrated endto-end off-road racing with a monocular camera by imitating a model predictive control expert with access to expensive sensors. In the context of urban driving, [36] showed that a similar concept can be applied to the driving simulator CARLA [12]. Driving simulators are an ideal proving ground for such approaches since they are inherently safe and can provide ground truth states.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

End-to-End Urban Driving by Imitating a Reinforcement Learning Coach

Zhang¹,

Liniger²,

Dai³

et al. 2021

2021 IEEE/CVF International Conference on Computer Vision (ICCV)

100

View full text Add to dashboard Cite

End-to-end approaches to autonomous driving commonly rely on expert demonstrations. Although humans are good drivers, they are not good coaches for end-to-end algorithms that demand dense on-policy supervision. On the contrary, automated experts that leverage privileged information can efficiently generate large scale on-policy and off-policy demonstrations. However, existing automated experts for urban driving make heavy use of hand-crafted rules and perform suboptimally even on driving simulators, where ground-truth information is available. To address these issues, we train a reinforcement learning expert that maps bird's-eye view images to continuous low-level actions. While setting a new performance upper-bound on CARLA, our expert is also a better coach that provides informative supervision signals for imitation learning agents to learn from. Supervised by our reinforcement learning coach, a baseline end-to-end agent with monocular camerainput achieves expert-level performance. Our end-to-end agent achieves a 78% success rate while generalizing to a new town and new weather on the NoCrash-dense benchmark and state-of-the-art performance on the more challenging CARLA LeaderBoard.

show abstract