Bootstrapping bilinear models of Simple Vehicles

Censi, Andrea; Murray, Richard M.

doi:10.1177/0278364914557708

Cited by 6 publications

(6 citation statements)

References 78 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our approach is related to visual servoing, which performs feedback control on features in an image [32], [33], [34] or the image pixels themselves [35]. Unlike visual servoing methods, our approach performs feedback control at the pixel level using a model learned entirely from unlabeled, real-world video data, without requiring any explicit camera calibration.…”

Section: Related Workmentioning

confidence: 99%

Deep visual foresight for planning robot motion

Finn

Levine

2017

2017 IEEE International Conference on Robotics and Automation (ICRA)

548

544

View full text Add to dashboard Cite

Abstract-A key challenge in scaling up robot learning to many skills and environments is removing the need for human supervision, so that robots can collect their own data and improve their own performance without being limited by the cost of requesting human feedback. Model-based reinforcement learning holds the promise of enabling an agent to learn to predict the effects of its actions, which could provide flexible predictive models for a wide range of tasks and environments, without detailed human supervision. We develop a method for combining deep action-conditioned video prediction models with model-predictive control that uses entirely unlabeled training data. Our approach does not require a calibrated camera, an instrumented training set-up, nor precise sensing and actuation. Our results show that our method enables a real robot to perform nonprehensile manipulation -pushing objects -and can handle novel objects not seen during training.

show abstract

Section: Related Workmentioning

confidence: 99%

Deep visual foresight for planning robot motion

Finn

Levine

2017

2017 IEEE International Conference on Robotics and Automation (ICRA)

548

544

View full text Add to dashboard Cite

show abstract

“…Their initial pilot program had six autonomous vehicles, modified Renault Zoes and Mitsubishi i-MiEVs, offering rides from predetermined pick-up and drop-off points within a 2.5 mi 2 radius. Since then, nuTonomy has partnered with a local ride-sharing company, Grab, to conduct further tests and plans to launch a commercial service in 2018 (Censi, 2017; Censi & Murray, 2015).…”

Section: Availability and Readiness Of Industry 40 Technologiesmentioning

confidence: 99%

Emerging Industrial Revolution: Symbiosis of Industry 4.0 and Circular Economy: The Role of Universities

Ramakrishna

Ngowi

Jager

et al. 2020

Science, Technology and Society

View full text Add to dashboard Cite

Growing consumerism and population worldwide raises concerns about society’s sustainability aspirations. This has led to calls for concerted efforts to shift from the linear economy to a circular economy (CE), which are gaining momentum globally. CE approaches lead to a zero-waste scenario of economic growth and sustainable development. These approaches are based on semi-scientific and empirical concepts with technologies enabling 3Rs (reduce, reuse, recycle) and 6Rs (reuse, recycle, redesign, remanufacture, reduce, recover). Studies estimate that the transition to a CE would save the world in excess of a trillion dollars annually while creating new jobs, business opportunities and economic growth. The emerging industrial revolution will enhance the symbiotic pursuit of new technologies and CE to transform extant production systems and business models for sustainability. This article examines the trends, availability and readiness of fourth industrial revolution (4IR or industry 4.0) technologies (for example, Internet of Things [IoT], artificial intelligence [AI] and nanotechnology) to support and promote CE transitions within the higher education institutional context. Furthermore, it elucidates the role of universities as living laboratories for experimenting the utility of industry 4.0 technologies in driving the shift towards CE futures. The article concludes that universities should play a pivotal role in engendering CE transitions.

show abstract

“…However, we observe that convolutional response maps can be interpreted as images and, under mild assumptions, the dynamics of image pixels during camera motion can be well approximated by means of a bilinear model (Censi & Murray, 2015). We therefore train a relatively simple bilinear model for short-term prediction of visual feature dynamics, which we can use inside a very simple visual servo that seeks to minimize the error between the next predicted feature values and a target image.…”

Section: Related Workmentioning

confidence: 99%

“…We use a bilinear model to represent these dynamics, motivated by prior work (Censi & Murray, 2015). In order to servo at different scales, we learn a bilinear dynamics model at each scale.…”

Section: Bilinear Dynamicsmentioning

confidence: 99%

Learning Visual Servoing with Deep Features and Fitted Q-Iteration

Lee,

Levine,

Abbeel

2017

Preprint

View full text Add to dashboard Cite

Visual servoing involves choosing actions that move a robot in response to observations from a camera, in order to reach a goal configuration in the world. Standard visual servoing approaches typically rely on manually designed features and analytical dynamics models, which limits their generalization capability and often requires extensive application-specific feature and model engineering.In this work, we study how learned visual features, learned predictive dynamics models, and reinforcement learning can be combined to learn visual servoing mechanisms. We focus on target following, with the goal of designing algorithms that can learn a visual servo using low amounts of data of the target in question, to enable quick adaptation to new targets. Our approach is based on servoing the camera in the space of learned visual features, rather than image pixels or manually-designed keypoints. We demonstrate that standard deep features, in our case taken from a model trained for object classification, can be used together with a bilinear predictive model to learn an effective visual servo that is robust to visual variation, changes in viewing angle and appearance, and occlusions. A key component of our approach is to use a sample-efficient fitted Q-iteration algorithm to learn which features are best suited for the task at hand. We show that we can learn an effective visual servo on a complex synthetic car following benchmark using just 20 training trajectory samples for reinforcement learning. We demonstrate substantial improvement over a conventional approach based on image pixels or hand-designed keypoints, and we show an improvement in sample-efficiency of more than two orders of magnitude over standard model-free deep reinforcement learning algorithms. Videos are available at http://rll.berkeley.edu/visual_servoing.

show abstract

Bootstrapping bilinear models of Simple Vehicles

Cited by 6 publications

References 78 publications

Deep visual foresight for planning robot motion

Deep visual foresight for planning robot motion

Emerging Industrial Revolution: Symbiosis of Industry 4.0 and Circular Economy: The Role of Universities

Learning Visual Servoing with Deep Features and Fitted Q-Iteration

Contact Info

Product

Resources

About