2019 International Conference on Robotics and Automation (ICRA) 2019
DOI: 10.1109/icra.2019.8794224
|View full text |Cite
|
Sign up to set email alerts
|

Deep Object-Centric Policies for Autonomous Driving

Abstract: While learning visuomotor skills in an end-toend manner is appealing, deep neural networks are often uninterpretable and fail in surprising ways. For robotics tasks, such as autonomous driving, models that explicitly represent objects may be more robust to new scenes and provide intuitive visualizations. We describe a taxonomy of "object-centric" models which leverage both object instances and end-to-end learning. In the Grand Theft Auto V simulator, we show that object-centric models outperform object-agnosti… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
2
1

Citation Types

0
80
0

Year Published

2019
2019
2023
2023

Publication Types

Select...
4
4
1

Relationship

1
8

Authors

Journals

citations
Cited by 90 publications
(80 citation statements)
references
References 22 publications
0
80
0
Order By: Relevance
“…The sub-scale vehicle successfully learned to drive at speeds up to 7.5m/s around the track. Instead of using direct vision for control, Wang et al [128] demonstrated that DAgger can be used to train an object-centric policy, which uses salient objects in the image (e.g. vehicles, pedestrians) to output a control action.…”
Section: Simultaneous Lateral and Longitudinal Control Systemsmentioning
confidence: 99%
“…The sub-scale vehicle successfully learned to drive at speeds up to 7.5m/s around the track. Instead of using direct vision for control, Wang et al [128] demonstrated that DAgger can be used to train an object-centric policy, which uses salient objects in the image (e.g. vehicles, pedestrians) to output a control action.…”
Section: Simultaneous Lateral and Longitudinal Control Systemsmentioning
confidence: 99%
“…[25] used an FCN-LSTM architecture with a segmentation mask to train a deep driving policy. [24] proposed an objectcentric model to predict the vehicle action with higher accuracy. Although both [25] and [24] achieved good prediction performance for complex urban scenarios, they did not provide closed loop evaluation either on real world or simulated environments.…”
Section: Related Workmentioning
confidence: 99%
“…[24] proposed an objectcentric model to predict the vehicle action with higher accuracy. Although both [25] and [24] achieved good prediction performance for complex urban scenarios, they did not provide closed loop evaluation either on real world or simulated environments. [22] used imitation learning to drive a simulated vehicle in closed loop, however it is restricted to limited scenarios such as lane following and lane changing with fixed number of surrounding vehicles.…”
Section: Related Workmentioning
confidence: 99%
“…However, they are not much efficient as one-stage networks. These methods are then applied to various fields, such as face recognition [17,18], semantic segmentation [19][20][21] and autonomous driving [22,23].…”
Section: Introductionmentioning
confidence: 99%