Stephen James

Younggyo Seo, Danijar Hafner, Hao Liu et al. · deepmind, utoronto

Visual model-based reinforcement learning (RL) has the potential to enable sample-efficient robot learning from visual observations. Yet the current approaches typically train a single model end-to-end for learning both visual representations and dynamics, making it difficult to accurately model the interaction between robots and small objects. In this work, we introduce a visual model-based RL framework that decouples visual representation learning and dynamics learning. Specifically, we train an autoencoder with convolutional layers and vision transformers (ViT) to reconstruct pixels given masked convolutional features, and learn a latent dynamics model that operates on the representations from the autoencoder. Moreover, to encode task-relevant information, we introduce an auxiliary reward prediction objective for the autoencoder. We continually update both autoencoder and dynamics model using online samples collected from environment interaction. We demonstrate that our decoupling approach achieves state-of-the-art performance on a variety of visual robotic tasks from Meta-world and RLBench, e.g., we achieve 81.7% success rate on 50 visual robotic manipulation tasks from Meta-world, while the baseline achieves 67.9%. Code is available on the project website: https://sites.google.com/view/mwm-rl.

Younggyo Seo, Kimin Lee, Stephen James et al.

Recent unsupervised pre-training methods have shown to be effective on language and vision domains by learning useful representations for multiple downstream tasks. In this paper, we investigate if such unsupervised pre-training methods can also be effective for vision-based reinforcement learning (RL). To this end, we introduce a framework that learns representations useful for understanding the dynamics via generative pre-training on videos. Our framework consists of two phases: we pre-train an action-free latent video prediction model, and then utilize the pre-trained representations for efficiently learning action-conditional world models on unseen environments. To incorporate additional action inputs during fine-tuning, we introduce a new architecture that stacks an action-conditional latent prediction model on top of the pre-trained action-free prediction model. Moreover, for better exploration, we propose a video-based intrinsic bonus that leverages pre-trained representations. We demonstrate that our framework significantly improves both final performances and sample-efficiency of vision-based RL in a variety of manipulation and locomotion tasks. Code is available at https://github.com/younggyoseo/apv.

Wilson Yan, Danijar Hafner, Stephen James et al. · deepmind, utoronto

To generate accurate videos, algorithms have to understand the spatial and temporal dependencies in the world. Current algorithms enable accurate predictions over short horizons but tend to suffer from temporal inconsistencies. When generated content goes out of view and is later revisited, the model invents different content instead. Despite this severe limitation, no established benchmarks on complex data exist for rigorously evaluating video generation with long temporal dependencies. In this paper, we curate 3 challenging video datasets with long-range dependencies by rendering walks through 3D scenes of procedural mazes, Minecraft worlds, and indoor scans. We perform a comprehensive evaluation of current models and observe their limitations in temporal consistency. Moreover, we introduce the Temporally Consistent Transformer (TECO), a generative model that substantially improves long-term consistency while also reducing sampling time. By compressing its input sequence into fewer embeddings, applying a temporal transformer, and expanding back using a spatial MaskGit, TECO outperforms existing models across many metrics. Videos are available on the website: https://wilson1yan.github.io/teco

Ilija Radosavovic, Tete Xiao, Stephen James et al.

In this work, we explore self-supervised visual pre-training on images from diverse, in-the-wild videos for real-world robotic tasks. Like prior work, our visual representations are pre-trained via a masked autoencoder (MAE), frozen, and then passed into a learnable control module. Unlike prior work, we show that the pre-trained representations are effective across a range of real-world robotic tasks and embodiments. We find that our encoder consistently outperforms CLIP (up to 75%), supervised ImageNet pre-training (up to 81%), and training from scratch (up to 81%). Finally, we train a 307M parameter vision transformer on a massive collection of 4.5M images from the Internet and egocentric videos, and demonstrate clearly the benefits of scaling visual pre-training for robot learning.

Younggyo Seo, Junsu Kim, Stephen James et al.

Visual robotic manipulation research and applications often use multiple cameras, or views, to better perceive the world. How else can we utilize the richness of multi-view data? In this paper, we investigate how to learn good representations with multi-view data and utilize them for visual robotic manipulation. Specifically, we train a multi-view masked autoencoder which reconstructs pixels of randomly masked viewpoints and then learn a world model operating on the representations from the autoencoder. We demonstrate the effectiveness of our method in a range of scenarios, including multi-view control and single-view control with auxiliary cameras for representation learning. We also show that the multi-view masked autoencoder trained with multiple randomized viewpoints enables training a policy with strong viewpoint randomization and transferring the policy to solve real-robot tasks without camera calibration and an adaptation procedure. Video demonstrations are available at: https://sites.google.com/view/mv-mwm.

Ademi Adeniji, Amber Xie, Carmelo Sferrazza et al.

Using learned reward functions (LRFs) as a means to solve sparse-reward reinforcement learning (RL) tasks has yielded some steady progress in task-complexity through the years. In this work, we question whether today's LRFs are best-suited as a direct replacement for task rewards. Instead, we propose leveraging the capabilities of LRFs as a pretraining signal for RL. Concretely, we propose $\textbf{LA}$nguage Reward $\textbf{M}$odulated $\textbf{P}$retraining (LAMP) which leverages the zero-shot capabilities of Vision-Language Models (VLMs) as a $\textit{pretraining}$ utility for RL as opposed to a downstream task reward. LAMP uses a frozen, pretrained VLM to scalably generate noisy, albeit shaped exploration rewards by computing the contrastive alignment between a highly diverse collection of language instructions and the image observations of an agent in its pretraining environment. LAMP optimizes these rewards in conjunction with standard novelty-seeking exploration rewards with reinforcement learning to acquire a language-conditioned, pretrained policy. Our VLM pretraining approach, which is a departure from previous attempts to use LRFs, can warmstart sample-efficient learning on robot manipulation tasks in RLBench.

Kai Chen, Rui Cao, Stephen James et al.

In this paper, we propose an iterative self-training framework for sim-to-real 6D object pose estimation to facilitate cost-effective robotic grasping. Given a bin-picking scenario, we establish a photo-realistic simulator to synthesize abundant virtual data, and use this to train an initial pose estimation network. This network then takes the role of a teacher model, which generates pose predictions for unlabeled real data. With these predictions, we further design a comprehensive adaptive selection scheme to distinguish reliable results, and leverage them as pseudo labels to update a student model for pose estimation on real data. To continuously improve the quality of pseudo labels, we iterate the above steps by taking the trained student model as a new teacher and re-label real data using the refined teacher model. We evaluate our method on a public benchmark and our newly-released dataset, achieving an ADD(-S) improvement of 11.49% and 22.62% respectively. Our method is also able to improve robotic bin-picking success by 19.54%, demonstrating the potential of iterative sim-to-real solutions for robotic applications.

Younggyo Seo, Kimin Lee, Fangchen Liu et al.

Video prediction is an important yet challenging problem; burdened with the tasks of generating future frames and learning environment dynamics. Recently, autoregressive latent video models have proved to be a powerful video prediction tool, by separating the video prediction into two sub-problems: pre-training an image generator model, followed by learning an autoregressive prediction model in the latent space of the image generator. However, successfully generating high-fidelity and high-resolution videos has yet to be seen. In this work, we investigate how to train an autoregressive latent video prediction model capable of predicting high-fidelity future frames with minimal modification to existing models, and produce high-resolution (256x256) videos. Specifically, we scale up prior models by employing a high-fidelity image generator (VQ-GAN) with a causal transformer model, and introduce additional techniques of top-k sampling and data augmentation to further improve video prediction quality. Despite the simplicity, the proposed method achieves competitive performance to state-of-the-art approaches on standard video prediction benchmarks with fewer parameters, and enables high-resolution video prediction on complex and large-scale datasets. Videos are available at https://sites.google.com/view/harp-videos/home.

Kai Chen, Stephen James, Congying Sui et al.

Most existing methods for category-level pose estimation rely on object point clouds. However, when considering transparent objects, depth cameras are usually not able to capture meaningful data, resulting in point clouds with severe artifacts. Without a high-quality point cloud, existing methods are not applicable to challenging transparent objects. To tackle this problem, we present StereoPose, a novel stereo image framework for category-level object pose estimation, ideally suited for transparent objects. For a robust estimation from pure stereo images, we develop a pipeline that decouples category-level pose estimation into object size estimation, initial pose estimation, and pose refinement. StereoPose then estimates object pose based on representation in the normalized object coordinate space~(NOCS). To address the issue of image content aliasing, we further define a back-view NOCS map for the transparent object. The back-view NOCS aims to reduce the network learning ambiguity caused by content aliasing, and leverage informative cues on the back of the transparent object for more accurate pose estimation. To further improve the performance of the stereo framework, StereoPose is equipped with a parallax attention module for stereo feature fusion and an epipolar loss for improving the stereo-view consistency of network predictions. Extensive experiments on the public TOD dataset demonstrate the superiority of the proposed StereoPose framework for category-level 6D transparent object pose estimation.

Stephen James, Pieter Abbeel

We propose Learned Path Ranking (LPR), a method that accepts an end-effector goal pose, and learns to rank a set of goal-reaching paths generated from an array of path generating methods, including: path planning, Bezier curve sampling, and a learned policy. The core idea being that each of the path generation modules will be useful in different tasks, or at different stages in a task. When LPR is added as an extension to C2F-ARM, our new system, C2F-ARM+LPR, retains the sample efficiency of its predecessor, while also being able to accomplish a larger set of tasks; in particular, tasks that require very specific motions (e.g. opening toilet seat) that need to be inferred from both demonstrations and exploration data. In addition to benchmarking our approach across 16 RLBench tasks, we also learn real-world tasks, tabula rasa, in 10-15 minutes, with only 3 demonstrations.

Amber Xie, Youngwoon Lee, Pieter Abbeel et al.

Contact is at the core of robotic manipulation. At times, it is desired (e.g. manipulation and grasping), and at times, it is harmful (e.g. when avoiding obstacles). However, traditional path planning algorithms focus solely on collision-free paths, limiting their applicability in contact-rich tasks. To address this limitation, we propose the domain of Language-Conditioned Path Planning, where contact-awareness is incorporated into the path planning problem. As a first step in this domain, we propose Language-Conditioned Collision Functions (LACO) a novel approach that learns a collision function using only a single-view image, language prompt, and robot configuration. LACO predicts collisions between the robot and the environment, enabling flexible, conditional path planning without the need for manual object annotations, point cloud data, or ground-truth object meshes. In both simulation and the real world, we demonstrate that LACO can facilitate complex, nuanced path plans that allow for interaction with objects that are safe to collide, rather than prohibiting any collision.

John So, Amber Xie, Sunggoo Jung et al.

Autonomous driving is complex, requiring sophisticated 3D scene understanding, localization, mapping, and control. Rather than explicitly modelling and fusing each of these components, we instead consider an end-to-end approach via reinforcement learning (RL). However, collecting exploration driving data in the real world is impractical and dangerous. While training in simulation and deploying visual sim-to-real techniques has worked well for robot manipulation, deploying beyond controlled workspace viewpoints remains a challenge. In this paper, we address this challenge by presenting Sim2Seg, a re-imagining of RCAN that crosses the visual reality gap for off-road autonomous driving, without using any real-world data. This is done by learning to translate randomized simulation images into simulated segmentation and depth maps, subsequently enabling real-world images to also be translated. This allows us to train an end-to-end RL policy in simulation, and directly deploy in the real-world. Our approach, which can be trained in 48 hours on 1 GPU, can perform equally as well as a classical perception and control stack that took thousands of engineering hours over several months to build. We hope this work motivates future end-to-end autonomous driving research.

Stephen James, Pieter Abbeel

Coarse-to-fine Q-attention enables sample-efficient robot manipulation by discretizing the translation space in a coarse-to-fine manner, where the resolution gradually increases at each layer in the hierarchy. Although effective, Q-attention suffers from "coarse ambiguity" - when voxelization is significantly coarse, it is not feasible to distinguish similar-looking objects without first inspecting at a finer resolution. To combat this, we propose to envision Q-attention as a tree that can be expanded and used to accumulate value estimates across the top-k voxels at each Q-attention depth. When our extension, Q-attention with Tree Expansion (QTE), replaces standard Q-attention in the Attention-driven Robot Manipulation (ARM) system, we are able to accomplish a larger set of tasks; especially on those that suffer from "coarse ambiguity". In addition to evaluating our approach across 12 RLBench tasks, we also show that the improved performance is visible in a real-world task involving small objects.

Wilson Yan, Ryo Okumura, Stephen James et al.

In this work, we present Patch-based Object-centric Video Transformer (POVT), a novel region-based video generation architecture that leverages object-centric information to efficiently model temporal dynamics in videos. We build upon prior work in video prediction via an autoregressive transformer over the discrete latent space of compressed videos, with an added modification to model object-centric information via bounding boxes. Due to better compressibility of object-centric representations, we can improve training efficiency by allowing the model to only access object information for longer horizon temporal information. When evaluated on various difficult object-centric datasets, our method achieves better or equal performance to other video generation models, while remaining computationally more efficient and scalable. In addition, we show that our method is able to perform object-centric controllability through bounding box manipulation, which may aid downstream tasks such as video editing, or visual planning. Samples are available at https://sites.google.com/view/povt-public

Jiangliu Wang, Jianbo Jiao, Yibing Song et al.

This work aims to improve unsupervised audio-visual pre-training. Inspired by the efficacy of data augmentation in visual contrastive learning, we propose a novel speed co-augmentation method that randomly changes the playback speeds of both audio and video data. Despite its simplicity, the speed co-augmentation method possesses two compelling attributes: (1) it increases the diversity of audio-visual pairs and doubles the size of negative pairs, resulting in a significant enhancement in the learned representations, and (2) it changes the strict correlation between audio-visual pairs but introduces a partial relationship between the augmented pairs, which is modeled by our proposed SoftInfoNCE loss to further boost the performance. Experimental results show that the proposed method significantly improves the learned representations when compared to vanilla audio-visual contrastive learning.

Nikita Chernyadev, Nicholas Backshall, Xiao Ma et al.

We introduce BiGym, a new benchmark and learning environment for mobile bi-manual demo-driven robotic manipulation. BiGym features 40 diverse tasks set in home environments, ranging from simple target reaching to complex kitchen cleaning. To capture the real-world performance accurately, we provide human-collected demonstrations for each task, reflecting the diverse modalities found in real-world robot trajectories. BiGym supports a variety of observations, including proprioceptive data and visual inputs such as RGB, and depth from 3 camera views. To validate the usability of BiGym, we thoroughly benchmark the state-of-the-art imitation learning algorithms and demo-driven reinforcement learning algorithms within the environment and discuss the future opportunities.

Younggyo Seo, Jafar Uruç, Stephen James

Despite recent advances in improving the sample-efficiency of reinforcement learning (RL) algorithms, designing an RL algorithm that can be practically deployed in real-world environments remains a challenge. In this paper, we present Coarse-to-fine Reinforcement Learning (CRL), a framework that trains RL agents to zoom-into a continuous action space in a coarse-to-fine manner, enabling the use of stable, sample-efficient value-based RL algorithms for fine-grained continuous control tasks. Our key idea is to train agents that output actions by iterating the procedure of (i) discretizing the continuous action space into multiple intervals and (ii) selecting the interval with the highest Q-value to further discretize at the next level. We then introduce a concrete, value-based algorithm within the CRL framework called Coarse-to-fine Q-Network (CQN). Our experiments demonstrate that CQN significantly outperforms RL and behavior cloning baselines on 20 sparsely-rewarded RLBench manipulation tasks with a modest number of environment interactions and expert demonstrations. We also show that CQN robustly learns to solve real-world manipulation tasks within a few minutes of online training.

Mohit Shridhar, Yat Long Lo, Stephen James

Image-generation diffusion models have been fine-tuned to unlock new capabilities such as image-editing and novel view synthesis. Can we similarly unlock image-generation models for visuomotor control? We present GENIMA, a behavior-cloning agent that fine-tunes Stable Diffusion to 'draw joint-actions' as targets on RGB images. These images are fed into a controller that maps the visual targets into a sequence of joint-positions. We study GENIMA on 25 RLBench and 9 real-world manipulation tasks. We find that, by lifting actions into image-space, internet pre-trained diffusion models can generate policies that outperform state-of-the-art visuomotor approaches, especially in robustness to scene perturbations and generalizing to novel objects. Our method is also competitive with 3D agents, despite lacking priors such as depth, keypoints, or motion-planners.

Zhao Mandi, Pieter Abbeel, Stephen James

Intelligent agents should have the ability to leverage knowledge from previously learned tasks in order to learn new ones quickly and efficiently. Meta-learning approaches have emerged as a popular solution to achieve this. However, meta-reinforcement learning (meta-RL) algorithms have thus far been restricted to simple environments with narrow task distributions. Moreover, the paradigm of pretraining followed by fine-tuning to adapt to new tasks has emerged as a simple yet effective solution in supervised and self-supervised learning. This calls into question the benefits of meta-learning approaches also in reinforcement learning, which typically come at the cost of high complexity. We hence investigate meta-RL approaches in a variety of vision-based benchmarks, including Procgen, RLBench, and Atari, where evaluations are made on completely novel tasks. Our findings show that when meta-learning approaches are evaluated on different tasks (rather than different variations of the same task), multi-task pretraining with fine-tuning on new tasks performs equally as well, or better, than meta-pretraining with meta test-time adaptation. This is encouraging for future research, as multi-task pretraining tends to be simpler and computationally cheaper than meta-RL. From these findings, we advocate for evaluating future meta-RL methods on more challenging tasks and including multi-task pretraining with fine-tuning as a simple, yet strong baseline.

Eugene Teoh, Sumit Patidar, Xiao Ma et al.

Generalising vision-based manipulation policies to novel environments remains a challenging area with limited exploration. Current practices involve collecting data in one location, training imitation learning or reinforcement learning policies with this data, and deploying the policy in the same location. However, this approach lacks scalability as it necessitates data collection in multiple locations for each task. This paper proposes a novel approach where data is collected in a location predominantly featuring green screens. We introduce Green-screen Augmentation (GreenAug), employing a chroma key algorithm to overlay background textures onto a green screen. Through extensive real-world empirical studies with over 850 training demonstrations and 8.2k evaluation episodes, we demonstrate that GreenAug surpasses no augmentation, standard computer vision augmentation, and prior generative augmentation methods in performance. While no algorithmic novelties are claimed, our paper advocates for a fundamental shift in data collection practices. We propose that real-world demonstrations in future research should utilise green screens, followed by the application of GreenAug. We believe GreenAug unlocks policy generalisation to visually distinct novel locations, addressing the current scene generalisation limitations in robot learning.

Shikun Liu, Stephen James, Andrew J. Davison et al.

Understanding the structure of multiple related tasks allows for multi-task learning to improve the generalisation ability of one or all of them. However, it usually requires training each pairwise combination of tasks together in order to capture task relationships, at an extremely high computational cost. In this work, we learn task relationships via an automated weighting framework, named Auto-Lambda. Unlike previous methods where task relationships are assumed to be fixed, Auto-Lambda is a gradient-based meta learning framework which explores continuous, dynamic task relationships via task-specific weightings, and can optimise any choice of combination of tasks through the formulation of a meta-loss; where the validation loss automatically influences task weightings throughout training. We apply the proposed framework to both multi-task and auxiliary learning problems in computer vision and robotics, and show that Auto-Lambda achieves state-of-the-art performance, even when compared to optimisation strategies designed specifically for each problem and data domain. Finally, we observe that Auto-Lambda can discover interesting learning behaviors, leading to new insights in multi-task learning. Code is available at https://github.com/lorenmt/auto-lambda.

Alexandru-Iosif Toma, Hussein Ali Jaafar, Hao-Ya Hsueh et al.

With the recent advances in machine learning, path planning algorithms are also evolving; however, the learned path planning algorithms often have difficulty competing with success rates of classic algorithms. We propose waypoint planning networks (WPN), a hybrid algorithm based on LSTMs with a local kernel - a classic algorithm such as A*, and a global kernel using a learned algorithm. WPN produces a more computationally efficient and robust solution. We compare WPN against A*, as well as related works including motion planning networks (MPNet) and value iteration networks (VIN). In this paper, the design and experiments have been conducted for 2D environments. Experimental results outline the benefits of WPN, both in efficiency and generalization. It is shown that WPN's search space is considerably less than A*, while being able to generate near optimal results. Additionally, WPN works on partial maps, unlike A* which needs the full map in advance. The code is available online.

Daniel Lenton, Stephen James, Ronald Clark et al.

Spatial memory, or the ability to remember and recall specific locations and objects, is central to autonomous agents' ability to carry out tasks in real environments. However, most existing artificial memory modules are not very adept at storing spatial information. We propose a parameter-free module, Egospheric Spatial Memory (ESM), which encodes the memory in an ego-sphere around the agent, enabling expressive 3D representations. ESM can be trained end-to-end via either imitation or reinforcement learning, and improves both training efficiency and final performance against other memory baselines on both drone and manipulator visuomotor control tasks. The explicit egocentric geometry also enables us to seamlessly combine the learned controller with other non-learned modalities, such as local obstacle avoidance. We further show applications to semantic segmentation on the ScanNet dataset, where ESM naturally combines image-level and map-level inference modalities. Through our broad set of experiments, we show that ESM provides a general computation graph for embodied spatial reasoning, and the module forms a bridge between real-time mapping systems and differentiable memory architectures. Implementation at: https://github.com/ivy-dl/memory.

Xiao Ma, Sumit Patidar, Iain Haughton et al.

This paper introduces Hierarchical Diffusion Policy (HDP), a hierarchical agent for multi-task robotic manipulation. HDP factorises a manipulation policy into a hierarchical structure: a high-level task-planning agent which predicts a distant next-best end-effector pose (NBP), and a low-level goal-conditioned diffusion policy which generates optimal motion trajectories. The factorised policy representation allows HDP to tackle both long-horizon task planning while generating fine-grained low-level actions. To generate context-aware motion trajectories while satisfying robot kinematics constraints, we present a novel kinematics-aware goal-conditioned control agent, Robot Kinematics Diffuser (RK-Diffuser). Specifically, RK-Diffuser learns to generate both the end-effector pose and joint position trajectories, and distill the accurate but kinematics-unaware end-effector pose diffuser to the kinematics-aware but less accurate joint position diffuser via differentiable kinematics. Empirically, we show that HDP achieves a significantly higher success rate than the state-of-the-art methods in both simulation and real-world.

Pietro Mazzaglia, Nicholas Backshall, Xiao Ma et al.

Joint space and task space control are the two dominant action modes for controlling robot arms within the robot learning literature. Actions in joint space provide precise control over the robot's pose, but tend to suffer from inefficient training; actions in task space boast data-efficient training but sacrifice the ability to perform tasks in confined spaces due to limited control over the full joint configuration. This work analyses the criteria for designing action spaces for robot manipulation and introduces ER (End-effector Redundancy), a novel action space formulation that, by addressing the redundancies present in the manipulator, aims to combine the advantages of both joint and task spaces, offering fine-grained comprehensive control with overactuated robot arms whilst achieving highly efficient robot learning. We present two implementations of ER, ERAngle (ERA) and ERJoint (ERJ), and we show that ERJ in particular demonstrates superior performance across multiple settings, especially when precise control over the robot configuration is required. We validate our results both in simulated and real robotic environments.

Kentaro Wada, Stephen James, Andrew J. Davison

Robots need the capability of placing objects in arbitrary, specific poses to rearrange the world and achieve various valuable tasks. Object reorientation plays a crucial role in this as objects may not initially be oriented such that the robot can grasp and then immediately place them in a specific goal pose. In this work, we present a vision-based manipulation system, ReorientBot, which consists of 1) visual scene understanding with pose estimation and volumetric reconstruction using an onboard RGB-D camera; 2) learned waypoint selection for successful and efficient motion generation for reorientation; 3) traditional motion planning to generate a collision-free trajectory from the selected waypoints. We evaluate our method using the YCB objects in both simulation and the real world, achieving 93% overall success, 81% improvement in success rate, and 22% improvement in execution time compared to a heuristic approach. We demonstrate extended multi-object rearrangement showing the general capability of the system.

Kentaro Wada, Stephen James, Andrew J. Davison

Robots need object-level scene understanding to manipulate objects while reasoning about contact, support, and occlusion among objects. Given a pile of objects, object recognition and reconstruction can identify the boundary of object instances, giving important cues as to how the objects form and support the pile. In this work, we present a system, SafePicking, that integrates object-level mapping and learning-based motion planning to generate a motion that safely extracts occluded target objects from a pile. Planning is done by learning a deep Q-network that receives observations of predicted poses and a depth-based heightmap to output a motion trajectory, trained to maximize a safety metric reward. Our results show that the observation fusion of poses and depth-sensing gives both better performance and robustness to the model. We evaluate our methods using the YCB objects in both simulation and the real world, achieving safe object extraction from piles.

Stephen James, Pieter Abbeel

We propose a new policy parameterization for representing 3D rotations during reinforcement learning. Today in the continuous control reinforcement learning literature, many stochastic policy parameterizations are Gaussian. We argue that universally applying a Gaussian policy parameterization is not always desirable for all environments. One such case in particular where this is true are tasks that involve predicting a 3D rotation output, either in isolation, or coupled with translation as part of a full 6D pose output. Our proposed Bingham Policy Parameterization (BPP) models the Bingham distribution and allows for better rotation (quaternion) prediction over a Gaussian policy parameterization in a range of reinforcement learning tasks. We evaluate BPP on the rotation Wahba problem task, as well as a set of vision-based next-best pose robot manipulation tasks from RLBench. We hope that this paper encourages more research into developing other policy parameterization that are more suited for particular environments, rather than always assuming Gaussian.

Stephen James, Kentaro Wada, Tristan Laidlow et al.

We present a coarse-to-fine discretisation method that enables the use of discrete reinforcement learning approaches in place of unstable and data-inefficient actor-critic methods in continuous robotics domains. This approach builds on the recently released ARM algorithm, which replaces the continuous next-best pose agent with a discrete one, with coarse-to-fine Q-attention. Given a voxelised scene, coarse-to-fine Q-attention learns what part of the scene to 'zoom' into. When this 'zooming' behaviour is applied iteratively, it results in a near-lossless discretisation of the translation space, and allows the use of a discrete action, deep Q-learning method. We show that our new coarse-to-fine algorithm achieves state-of-the-art performance on several difficult sparsely rewarded RLBench vision-based robotics tasks, and can train real-world policies, tabula rasa, in a matter of minutes, with as little as 3 demonstrations.

Stephen James, Andrew J. Davison

Despite the success of reinforcement learning methods, they have yet to have their breakthrough moment when applied to a broad range of robotic manipulation tasks. This is partly due to the fact that reinforcement learning algorithms are notoriously difficult and time consuming to train, which is exacerbated when training from images rather than full-state inputs. As humans perform manipulation tasks, our eyes closely monitor every step of the process with our gaze focusing sequentially on the objects being manipulated. With this in mind, we present our Attention-driven Robotic Manipulation (ARM) algorithm, which is a general manipulation algorithm that can be applied to a range of sparse-rewarded tasks, given only a small number of demonstrations. ARM splits the complex task of manipulation into a 3 stage pipeline: (1) a Q-attention agent extracts relevant pixel locations from RGB and point cloud inputs, (2) a next-best pose agent that accepts crops from the Q-attention agent and outputs poses, and (3) a control agent that takes the goal pose and outputs joint actions. We show that current learning algorithms fail on a range of RLBench tasks, whilst ARM is successful.

Zoe Landgraf, Raluca Scona, Tristan Laidlow et al.

By estimating 3D shape and instances from a single view, we can capture information about an environment quickly, without the need for comprehensive scanning and multi-view fusion. Solving this task for composite scenes (such as object stacks) is challenging: occluded areas are not only ambiguous in shape but also in instance segmentation; multiple decompositions could be valid. We observe that physics constrains decomposition as well as shape in occluded regions and hypothesise that a latent space learned from scenes built under physics simulation can serve as a prior to better predict shape and instances in occluded regions. To this end we propose SIMstack, a depth-conditioned Variational Auto-Encoder (VAE), trained on a dataset of objects stacked under physics simulation. We formulate instance segmentation as a centre voting task which allows for class-agnostic detection and doesn't require setting the maximum number of objects in the scene. At test time, our model can generate 3D shape and instance segmentation from a single depth view, probabilistically sampling proposals for the occluded region from the learned latent space. Our method has practical applications in providing robots some of the ability humans have to make rapid intuitive inferences of partially observed scenes. We demonstrate an application for precise (non-disruptive) object grasping of unknown objects from a single depth view.

Daniel Lenton, Fabio Pardo, Fabian Falck et al.

We introduce Ivy, a templated Deep Learning (DL) framework which abstracts existing DL frameworks. Ivy unifies the core functions of these frameworks to exhibit consistent call signatures, syntax and input-output behaviour. New high-level framework-agnostic functions and classes, which are usable alongside framework-specific code, can then be implemented as compositions of the unified low-level Ivy functions. Ivy currently supports TensorFlow, PyTorch, MXNet, Jax and NumPy. We also release four pure-Ivy libraries for mechanics, 3D vision, robotics, and differentiable environments. Through our evaluations, we show that Ivy can significantly reduce lines of code with a runtime overhead of less than 1% in most cases. We welcome developers to join the Ivy community by writing their own functions, layers and libraries in Ivy, maximizing their audience and helping to accelerate DL research through inter-framework codebases. More information can be found at https://ivy-dl.org.

Kentaro Wada, Edgar Sucar, Stephen James et al.

Robots and other smart devices need efficient object-based scene representations from their on-board vision systems to reason about contact, physics and occlusion. Recognized precise object models will play an important role alongside non-parametric reconstructions of unrecognized structures. We present a system which can estimate the accurate poses of multiple known objects in contact and occlusion from real-time, embodied multi-view vision. Our approach makes 3D object pose proposals from single RGB-D views, accumulates pose estimates and non-parametric occupancy information from multiple views as the camera moves, and performs joint optimization to estimate consistent, non-intersecting poses for multiple objects in contact. We verify the accuracy and robustness of our approach experimentally on 2 object datasets: YCB-Video, and our own challenging Cluttered YCB-Video. We demonstrate a real-time robotics application where a robot arm precisely and orderly disassembles complicated piles of objects, using only on-board RGB-D vision.

Alessandro Bonardi, Stephen James, Andrew J. Davison

Humans can naturally learn to execute a new task by seeing it performed by other individuals once, and then reproduce it in a variety of configurations. Endowing robots with this ability of imitating humans from third person is a very immediate and natural way of teaching new tasks. Only recently, through meta-learning, there have been successful attempts to one-shot imitation learning from humans; however, these approaches require a lot of human resources to collect the data in the real world to train the robot. But is there a way to remove the need for real world human demonstrations during training? We show that with Task-Embedded Control Networks, we can infer control polices by embedding human demonstrations that can condition a control policy and achieve one-shot imitation learning. Importantly, we do not use a real human arm to supply demonstrations during training, but instead leverage domain randomisation in an application that has not been seen before: sim-to-real transfer on humans. Upon evaluating our approach on pushing and placing tasks in both simulation and in the real world, we show that in comparison to a system that was trained on real-world data we are able to achieve similar results by utilising only simulation data.

Stephen James, Zicong Ma, David Rovick Arrojo et al.

We present a challenging new benchmark and learning-environment for robot learning: RLBench. The benchmark features 100 completely unique, hand-designed tasks ranging in difficulty, from simple target reaching and door opening, to longer multi-stage tasks, such as opening an oven and placing a tray in it. We provide an array of both proprioceptive observations and visual observations, which include rgb, depth, and segmentation masks from an over-the-shoulder stereo camera and an eye-in-hand monocular camera. Uniquely, each task comes with an infinite supply of demos through the use of motion planners operating on a series of waypoints given during task creation time; enabling an exciting flurry of demonstration-based learning. RLBench has been designed with scalability in mind; new tasks, along with their motion-planned demos, can be easily created and then verified by a series of tools, allowing users to submit their own tasks to the RLBench task repository. This large-scale benchmark aims to accelerate progress in a number of vision-guided manipulation research areas, including: reinforcement learning, imitation learning, multi-task learning, geometric computer vision, and in particular, few-shot learning. With the benchmark's breadth of tasks and demonstrations, we propose the first large-scale few-shot challenge in robotics. We hope that the scale and diversity of RLBench offers unparalleled research opportunities in the robot learning community and beyond.

Stephen James, Marc Freese, Andrew J. Davison

PyRep is a toolkit for robot learning research, built on top of the virtual robotics experimentation platform (V-REP). Through a series of modifications and additions, we have created a tailored version of V-REP built with robot learning in mind. The new PyRep toolkit offers three improvements: (1) a simple and flexible API for robot control and scene manipulation, (2) a new rendering engine, and (3) speed boosts upwards of 10,000x in comparison to the previous Python Remote API. With these improvements, we believe PyRep is the ideal toolkit to facilitate rapid prototyping of learning algorithms in the areas of reinforcement learning, imitation learning, state estimation, mapping, and computer vision.

Stephen James, Paul Wohlhart, Mrinal Kalakrishnan et al.

Real world data, especially in the domain of robotics, is notoriously costly to collect. One way to circumvent this can be to leverage the power of simulation to produce large amounts of labelled data. However, training models on simulated images does not readily transfer to real-world ones. Using domain adaptation methods to cross this "reality gap" requires a large amount of unlabelled real-world data, whilst domain randomization alone can waste modeling power. In this paper, we present Randomized-to-Canonical Adaptation Networks (RCANs), a novel approach to crossing the visual reality gap that uses no real-world data. Our method learns to translate randomized rendered images into their equivalent non-randomized, canonical versions. This in turn allows for real images to also be translated into canonical sim images. We demonstrate the effectiveness of this sim-to-real approach by training a vision-based closed-loop grasping reinforcement learning agent in simulation, and then transferring it to the real world to attain 70% zero-shot grasp success on unseen objects, a result that almost doubles the success of learning the same task directly on domain randomization alone. Additionally, by joint finetuning in the real-world with only 5,000 real-world grasps, our method achieves 91%, attaining comparable performance to a state-of-the-art system trained with 580,000 real-world grasps, resulting in a reduction of real-world data by more than 99%.

Stephen James, Michael Bloesch, Andrew J. Davison

Much like humans, robots should have the ability to leverage knowledge from previously learned tasks in order to learn new tasks quickly in new and unfamiliar environments. Despite this, most robot learning approaches have focused on learning a single task, from scratch, with a limited notion of generalisation, and no way of leveraging the knowledge to learn other tasks more efficiently. One possible solution is meta-learning, but many of the related approaches are limited in their ability to scale to a large number of tasks and to learn further tasks without forgetting previously learned ones. With this in mind, we introduce Task-Embedded Control Networks, which employ ideas from metric learning in order to create a task embedding that can be used by a robot to learn new tasks from one or more demonstrations. In the area of visually-guided manipulation, we present simulation results in which we surpass the performance of a state-of-the-art method when using only visual information from each demonstration. Additionally, we demonstrate that our approach can also be used in conjunction with domain randomisation to train our few-shot learning ability in simulation and then deploy in the real world without any additional training. Once deployed, the robot can learn new tasks from a single real-world demonstration.

Jan Matas, Stephen James, Andrew J. Davison

We have seen much recent progress in rigid object manipulation, but interaction with deformable objects has notably lagged behind. Due to the large configuration space of deformable objects, solutions using traditional modelling approaches require significant engineering work. Perhaps then, bypassing the need for explicit modelling and instead learning the control in an end-to-end manner serves as a better approach? Despite the growing interest in the use of end-to-end robot learning approaches, only a small amount of work has focused on their applicability to deformable object manipulation. Moreover, due to the large amount of data needed to learn these end-to-end solutions, an emerging trend is to learn control policies in simulation and then transfer them over to the real world. To-date, no work has explored whether it is possible to learn and transfer deformable object policies. We believe that if sim-to-real methods are to be employed further, then it should be possible to learn to interact with a wide variety of objects, and not only rigid objects. In this work, we use a combination of state-of-the-art deep reinforcement learning algorithms to solve the problem of manipulating deformable objects (specifically cloth). We evaluate our approach on three tasks --- folding a towel up to a mark, folding a face towel diagonally, and draping a piece of cloth over a hanger. Our agents are fully trained in simulation with domain randomisation, and then successfully deployed in the real world without having seen any real deformable objects.

Stephen James, Andrew J. Davison, Edward Johns

End-to-end control for robot manipulation and grasping is emerging as an attractive alternative to traditional pipelined approaches. However, end-to-end methods tend to either be slow to train, exhibit little or no generalisability, or lack the ability to accomplish long-horizon or multi-stage tasks. In this paper, we show how two simple techniques can lead to end-to-end (image to velocity) execution of a multi-stage task, which is analogous to a simple tidying routine, without having seen a single real image. This involves locating, reaching for, and grasping a cube, then locating a basket and dropping the cube inside. To achieve this, robot trajectories are computed in a simulator, to collect a series of control velocities which accomplish the task. Then, a CNN is trained to map observed images to velocities, using domain randomisation to enable generalisation to real world images. Results show that we are able to successfully accomplish the task in the real world with the ability to generalise to novel environments, including those with dynamic lighting conditions, distractor objects, and moving objects, including the basket itself. We believe our approach to be simple, highly scalable, and capable of learning long-horizon tasks that have until now not been shown with the state-of-the-art in end-to-end robot control.

Stephen James, Edward Johns

Recent trends in robot arm control have seen a shift towards end-to-end solutions, using deep reinforcement learning to learn a controller directly from raw sensor data, rather than relying on a hand-crafted, modular pipeline. However, the high dimensionality of the state space often means that it is impractical to generate sufficient training data with real-world experiments. As an alternative solution, we propose to learn a robot controller in simulation, with the potential of then transferring this to a real robot. Building upon the recent success of deep Q-networks, we present an approach which uses 3D simulations to train a 7-DOF robotic arm in a control task without any prior knowledge. The controller accepts images of the environment as its only input, and outputs motor actions for the task of locating and grasping a cube, over a range of initial configurations. To encourage efficient learning, a structured reward function is designed with intermediate rewards. We also present preliminary results in direct transfer of policies over to a real robot, without any further training.