Reinforcement learning (RL) is a framework that lets agents learn decision making from experience. One of the many variants of RL is off-policy RL, where an agent is trained using a combination of data collected by other agents (off-policy data) and data it collects itself to learn generalizable skills like robotic walking and grasping. In contrast, fully off-policy RL is a variant in which an agent learns entirely from older data, which is appealing because it enables model iteration without requiring a physical robot. With fully off-policy RL, one can train several models on the same fixed dataset collected by previous agents, then select the best one. However, fully off-policy RL comes with a catch: while training can occur without a real robot, evaluation of the models cannot. Furthermore, ground-truth evaluation with a physical robot is too inefficient to test promising approaches that require evaluating a large number of models, such as automated architecture search with AutoML.

This challenge motivates off-policy evaluation (OPE), techniques for studying the quality of new agents using data from other agents. With rankings from OPE, we can selectively test only the most promising models on real-world robots, significantly scaling experimentation with the same fixed real robot budget.

A diagram for real-world model development. Assuming we can evaluate 10 models per day, without off-policy evaluation, we would need 100x as many days to evaluate our models.

Though the OPE framework shows promise, it assumes one has an off-policy evaluation method that accurately ranks performance from old data. However, agents that collected past experience may act very differently from newer learned agents, which makes it hard to get good estimates of performance.

In “Off-Policy Evaluation via Off-Policy Classification”, we propose a new off-policy evaluation method, called off-policy classification (OPC), that evaluates the performance of agents from past data by treating evaluation as a classification problem, in which actions are labeled as either potentially leading to success or guaranteed to result in failure. Our method works for image (camera) inputs, and doesn’t require reweighting data with importance sampling or using accurate models of the target environment, two approaches commonly used in prior work. We show that OPC scales to larger tasks, including a vision-based robotic grasping task in the real world.

How OPC Works
OPC relies on two assumptions: 1) that the final task has deterministic dynamics, i.e. no randomness is involved in how states change, and 2) that the agent either succeeds or fails at the end of each trial. This second “success or failure” assumption is natural for many tasks, such as picking up an object, solving a maze, winning a game, and so on. Because each trial will either succeed or fail in a deterministic way, we can assign binary classification labels to each action. We say an action is effective if it could lead to success, and is catastrophic if it is guaranteed to lead to failure.

OPC utilizes a Q-function, learned with a Q-learning algorithm, that estimates the future total reward if the agent chooses to take some action from its current state. The agent will then choose the action with the largest total reward estimate. In our paper, we prove that the performance of an agent is measured by how often its chosen action is an effective action, which depends on how well the Q-function correctly classifies actions as effective vs. catastrophic. This classification accuracy acts as an off-policy evaluation score.

However, the labeling of data from previous trials is only partial. For example, if a previous trial was a failure, we do not get negative labels because we do not know which action was the catastrophic one. To overcome this, we leverage techniques from semi-supervised learning, positive-unlabeled learning in particular, to get an estimate of classification accuracy from partially labeled data. This accuracy is the OPC score.

Off-Policy Evaluation for Sim-to-Real Learning
In robotics, it’s common to use simulated data and transfer learning techniques to reduce the sample complexity of learning robotics skills. This can be very useful, but tuning these sim-to-real techniques for real-world robotics is challenging. Much like off-policy RL, training doesn’t use the real robot, because it is trained in simulation, but evaluation of that policy still needs to use a real robot. Here, off-policy evaluation can come to the rescue again—we can take a policy trained only in simulation, then evaluate it using previous real-world data to measure its transfer to the real robot. We examine OPC across both fully off-policy RL and sim-to-real RL.

An example of how simulated experience can differ from real-world experience. Here, simulated images (left) have much less visual complexity than real-world images (right).

Results
First, we set up a simulated version of our robot grasping task, where we could easily train and evaluate several models to benchmark off-policy evaluation. These models were trained with fully off-policy RL, then evaluated with off-policy evaluation. We found that in our robotics tasks, a variant of the OPC called the SoftOPC performed best at predicting final success rate.

An experiment in the simulated grasping task. The red curve is the dimensionless SoftOPC score over the course of training, evaluated from old data. The blue curve is the grasp success rate in simulation. We see the SoftOPC on old data correlates well with grasp success of the model within our simulator.

After success in sim, we then tried SoftOPC in the real-world task. We took 15 models, trained to have varying degrees of robustness to the gap between simulation and reality. Of these models, 7 of them were trained purely in simulation, and the rest were trained on mixes of simulated and real-world data. For each model, we evaluated the SoftOPC on off-policy real-world data, then the real-world grasp success, to see how well SoftOPC predicted performance of that model. We found that on real data, the SoftOPC does produce scores that correlate with true grasp success, letting us rank sim-to-real techniques using past real experience.

SoftOPC score and true performance for 3 different sim-to-real methods: a baseline simulation, a simulation with random textures and lighting, and a model trained with RCAN. All three models are trained with no real data, then evaluated with off-policy evaluation on a validation set of real data. The ordering of the SoftOPC score matches the order of real grasp success.

Below is a scatterplot of the full results from all 15 models. Each point represents the off-policy evaluation score and real-world grasp success of each model. We compare different scoring functions by their correlation to final grasp success. The SoftOPC does not correlate perfectly with true grasp success, but its scores are significantly more reliable than baseline approaches like the temporal-difference error (the standard Q-learning loss).

Results from our sim-to-real evaluation experiment. On the left is a baseline, the temporal difference error of the model. On the right is one of our proposed methods, the SoftOPC. The shaded region is a 95% confidence interval. The correlation is significantly better with SoftOPC.

Future Work
One promising direction for future work is to see if we can relax our assumptions about the task, to support tasks where dynamics are more noisy, or where we get partial credit for almost succeeding. However, even with our included assumptions, we think the results are promising enough to be applied to many real-world RL problems.

Acknowledgements
This research was conducted by Alex Irpan, Kanishka Rao, Konstantinos Bousmalis, Chris Harris, Julian Ibarz and Sergey Levine. We’d like to thank Razvan Pascanu, Dale Schuurmans, George Tucker and Paul Wohlhart for valuable discussions. A preprint is available on arXiv.

This week, Long Beach, CA hosts the 2019 Conference on Computer Vision and Pattern Recognition (CVPR 2019), the premier annual computer vision event comprising the main conference and several co-located workshops and tutorials. As a leader in computer vision research and a Platinum Sponsor, Google will have a strong presence at CVPR 2019—over 250 Googlers will be in attendance to present papers and invited talks at the conference, and to organize and participate in multiple workshops.

If you are attending CVPR this year, please stop by our booth and chat with our researchers who are actively pursuing the next generation of intelligent systems that utilize the latest machine learning techniques applied to various areas of machine perception. Our researchers will also be available to talk about and demo several recent efforts, including the technology behind predicting pedestrian motion, the Open Images V5 dataset and much more.

You can learn more about our research being presented at CVPR 2019 in the list below (Google affiliations highlighted in blue)

Area Chairs include:
Jonathan T. Barron, William T. Freeman, Ce Liu, Michael Ryoo, Noah Snavely

Oral Presentations
Relational Action Forecasting
Chen Sun, Abhinav Shrivastava, Carl Vondrick, Rahul Sukthankar, Kevin Murphy, Cordelia Schmid

Pushing the Boundaries of View Extrapolation With Multiplane Images
Pratul P. Srinivasan, Richard Tucker, Jonathan T. Barron, Ravi Ramamoorthi, Ren Ng, Noah Snavely

Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation
Chenxi Liu, Liang-Chieh Chen, Florian Schroff, Hartwig Adam, Wei Hua, Alan L. Yuille, Li Fei-Fei

AutoAugment: Learning Augmentation Strategies From Data
Ekin D. Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasudevan, Quoc V. Le

DeepView: View Synthesis With Learned Gradient Descent
John Flynn, Michael Broxton, Paul Debevec, Matthew DuVall, Graham Fyffe, Ryan Overbeck, Noah Snavely, Richard Tucker

Normalized Object Coordinate Space for Category-Level 6D Object Pose and Size Estimation
He Wang, Srinath Sridhar, Jingwei Huang, Julien Valentin, Shuran Song, Leonidas J. Guibas

Do Better ImageNet Models Transfer Better?
Simon Kornblith, Jonathon Shlens, Quoc V. Le

TextureNet: Consistent Local Parametrizations for Learning From High-Resolution Signals on Meshes
Jingwei Huang, Haotian Zhang, Li Yi, Thomas Funkhouser, Matthias Niessner, Leonidas J. Guibas

Diverse Generation for Multi-Agent Sports Games
Raymond A. Yeh, Alexander G. Schwing, Jonathan Huang, Kevin Murphy

Occupancy Networks: Learning 3D Reconstruction in Function Space
Lars Mescheder, Michael Oechsle, Michael Niemeyer, Sebastian Nowozin, Andreas Geiger

A General and Adaptive Robust Loss Function
Jonathan T. Barron

Learning the Depths of Moving People by Watching Frozen People
Zhengqi Li, Tali Dekel, Forrester Cole, Richard Tucker, Noah Snavely, Ce Liu, William T. Freeman

Composing Text and Image for Image Retrieval – an Empirical Odyssey
Nam Vo, Lu Jiang, Chen Sun, Kevin Murphy, Li-Jia Li, Li Fei-Fei, James Hays

Learning to Synthesize Motion Blur
Tim Brooks, Jonathan T. Barron

Neural Rerendering in the Wild
Moustafa Meshry, Dan B. Goldman, Sameh Khamis, Hugues Hoppe, Rohit Pandey, Noah Snavely, Ricardo Martin-Brualla

Neural Illumination: Lighting Prediction for Indoor Environments
Shuran Song, Thomas Funkhouser

Unprocessing Images for Learned Raw Denoising
Tim Brooks, Ben Mildenhall, Tianfan Xue, Jiawen Chen, Dillon Sharlet, Jonathan T. Barron

Posters
Co-Occurrent Features in Semantic Segmentation
Hang Zhang, Han Zhang, Chenguang Wang, Junyuan Xie

CrDoCo: Pixel-Level Domain Transfer With Cross-Domain Consistency
Yun-Chun Chen, Yen-Yu Lin, Ming-Hsuan Yang, Jia-Bin Huang

Im2Pencil: Controllable Pencil Illustration From Photographs
Yijun Li, Chen Fang, Aaron Hertzmann, Eli Shechtman, Ming-Hsuan Yang

Mode Seeking Generative Adversarial Networks for Diverse Image Synthesis
Qi Mao, Hsin-Ying Lee, Hung-Yu Tseng, Siwei Ma, Ming-Hsuan Yang

Revisiting Self-Supervised Visual Representation Learning
Alexander Kolesnikov, Xiaohua Zhai, Lucas Beyer

Scene Graph Generation With External Knowledge and Image Reconstruction
Jiuxiang Gu, Handong Zhao, Zhe Lin, Sheng Li, Jianfei Cai, Mingyang Ling

Scene Memory Transformer for Embodied Agents in Long-Horizon Tasks
Kuan Fang, Alexander Toshev, Li Fei-Fei, Silvio Savarese

Spatially Variant Linear Representation Models for Joint Filtering
Jinshan Pan, Jiangxin Dong, Jimmy S. Ren, Liang Lin, Jinhui Tang, Ming-Hsuan Yang

Target-Aware Deep Tracking
Xin Li, Chao Ma, Baoyuan Wu, Zhenyu He, Ming-Hsuan Yang

Temporal Cycle-Consistency Learning
Debidatta Dwibedi, Yusuf Aytar, Jonathan Tompson, Pierre Sermanet, Andrew Zisserman

Depth-Aware Video Frame Interpolation
Wenbo Bao, Wei-Sheng Lai, Chao Ma, Xiaoyun Zhang, Zhiyong Gao, Ming-Hsuan Yang

MnasNet: Platform-Aware Neural Architecture Search for Mobile
Mingxing Tan, Bo Chen, Ruoming Pang, Vijay Vasudevan, Mark Sandler, Andrew Howard, Quoc V. Le

A Compact Embedding for Facial Expression Similarity
Raviteja Vemulapalli, Aseem Agarwala

Contrastive Adaptation Network for Unsupervised Domain Adaptation
Guoliang Kang, Lu Jiang, Yi Yang, Alexander G. Hauptmann

DeepLight: Learning Illumination for Unconstrained Mobile Mixed Reality
Chloe LeGendre, Wan-Chun Ma, Graham Fyffe, John Flynn, Laurent Charbonnel, Jay Busch, Paul Debevec

Detect-To-Retrieve: Efficient Regional Aggregation for Image Search
Marvin Teichmann, Andre Araujo, Menglong Zhu, Jack Sim

Fast Object Class Labelling via Speech
Michael Gygli, Vittorio Ferrari

Learning Independent Object Motion From Unlabelled Stereoscopic Videos
Zhe Cao, Abhishek Kar, Christian Hane, Jitendra Malik

Peeking Into the Future: Predicting Future Person Activities and Locations in Videos
Junwei Liang, Lu Jiang, Juan Carlos Niebles, Alexander G. Hauptmann, Li Fei-Fei

SpotTune: Transfer Learning Through Adaptive Fine-Tuning
Yunhui Guo, Honghui Shi, Abhishek Kumar, Kristen Grauman, Tajana Rosing, Rogerio Feris

NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection
Golnaz Ghiasi, Tsung-Yi Lin, Quoc V. Le

Class-Balanced Loss Based on Effective Number of Samples
Yin Cui, Menglin Jia, Tsung-Yi Lin, Yang Song, Serge Belongie

FEELVOS: Fast End-To-End Embedding Learning for Video Object Segmentation
Paul Voigtlaender, Yuning Chai, Florian Schroff, Hartwig Adam, Bastian Leibe, Liang-Chieh Chen

Inserting Videos Into Videos
Donghoon Lee, Tomas Pfister, Ming-Hsuan Yang

Volumetric Capture of Humans With a Single RGBD Camera via Semi-Parametric Learning
Rohit Pandey, Anastasia Tkach, Shuoran Yang, Pavel Pidlypenskyi, Jonathan Taylor, Ricardo Martin-Brualla, Andrea Tagliasacchi, George Papandreou, Philip Davidson, Cem Keskin, Shahram Izadi, Sean Fanello

You Look Twice: GaterNet for Dynamic Filter Selection in CNNs
Zhourong Chen, Yang Li, Samy Bengio, Si Si

Interactive Full Image Segmentation by Considering All Regions Jointly
Eirikur Agustsson, Jasper R. R. Uijlings, Vittorio Ferrari

Large-Scale Interactive Object Segmentation With Human Annotators
Rodrigo Benenson, Stefan Popov, Vittorio Ferrari

Self-Supervised GANs via Auxiliary Rotation Loss
Ting Chen, Xiaohua Zhai, Marvin Ritter, Mario Lučić, Neil Houlsby

Sim-To-Real via Sim-To-Sim: Data-Efficient Robotic Grasping via Randomized-To-Canonical Adaptation Networks
Stephen James, Paul Wohlhart, Mrinal Kalakrishnan, Dmitry Kalashnikov, Alex Irpan, Julian Ibarz, Sergey Levine, Raia Hadsell, Konstantinos Bousmalis

Using Unknown Occluders to Recover Hidden Scenes
Adam B. Yedidia, Manel Baradad, Christos Thrampoulidis, William T. Freeman, Gregory W. Wornell

Workshops
Computer Vision for Global Challenges
Organizers include: Timnit Gebru, Ernest Mwebaze, John Quinn

Deep Vision 2019
Invited speakers include: Pierre Sermanet, Chris Bregler

Landmark Recognition
Organizers include: Andre Araujo, Bingyi Cao, Jack Sim, Tobias Weyand

Image Matching: Local Features and Beyond
Organizers include: Eduard Trulls

3D-WiDGET: Deep GEneraTive Models for 3D Understanding
Invited speakers include: Julien Valentin

Fine-Grained Visual Categorization
Organizers include: Christine Kaeser-Chen
Advisory panel includes: Hartwig Adam

Low-Power Image Recognition Challenge (LPIRC)
Organizers include: Aakanksha Chowdhery, Achille Brighton, Alec Go, Andrew Howard, Bo Chen, Jaeyoun Kim, Jeff Gilbert

New Trends in Image Restoration and Enhancement Workshop and Associated Challenges
Program chairs include: Vivek Kwatra, Peyman Milanfar, Sebastian Nowozin, George Toderici, Ming-Hsuan Yang

Spatio-temporal Action Recognition (AVA) @ ActivityNet Challenge
Organizers include: David Ross, Sourish Chaudhuri, Radhika Marvin, Arkadiusz Stopczynski, Joseph Roth, Caroline Pantofaru, Chen Sun, Cordelia Schmid

Third Workshop on Computer Vision for AR/VR
Organizers include: Sofien Bouaziz, Serge Belongie

DAVIS Challenge on Video Object Segmentation
Organizers include: Jordi Pont-Tuset, Alberto Montes

Efficient Deep Learning for Computer Vision
Invited speakers include: Andrew Howard

Fairness Accountability Transparency and Ethics in Computer Vision
Organizers include: Timnit Gebru, Margaret Mitchell

Precognition Seeing through the Future
Organizers include: Utsav Prabhu

Workshop and Challenge on Learned Image Compression
Organizers include: George Toderici, Michele Covell, Johannes Ballé, Eirikur Agustsson, Nick Johnston

When Blockchain Meets Computer Vision & AI
Invited speakers include: Chris Bregler

Applications of Computer Vision and Pattern Recognition to Media Forensics
Organizers include: Paul Natsev, Christoph Bregler

Tutorials
Towards Relightable Volumetric Performance Capture of Humans
Organizers include: Sean Fanello, Christoph Rhemann, Graham Fyffe, Jonathan Taylor, Sofien Bouaziz, Paul Debevec, Shahram Izadi

Learning Representations via Graph-structured Networks
Organizers include: Ming-Hsuan Yang