RLBench: The Robot Learning Benchmark & Learning Environment

By Stephen James, Zicong Ma, David Rovick Arrojo, Andrew J. Davison. 2019

985 citations recorded by the source corpus.

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 possibilities. 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. Benchmarking code and videos can be found at https://sites.google.com/view/rlbench.

Subfields: Artificial Intelligence, Control and Systems Engineering, Evaluation & Benchmarks, Robotics & Control

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Control and Systems Engineering

In Evaluation & Benchmarks

In Robotics & Control

Citations in the corpus

Habitat 2.0: Training Home Assistants to Rearrange their Habitat (2021, background): “SAPIEN [35] OpenGL/OptiX configurable PhysX rigid/articulated dynamics object-level 200 - 400† RLBench [36] CoppeliaSim (OpenGL) Gouraud shading CoppeliaSim (Bullet/ODE) rigid/articulated dynamics table-top 1 - 60† iGibson [37] PyRender PBR shading PyBullet rigid/articulated dynamics house-scale 100”
CausalWorld: A Robotic Manipulation Benchmark for Causal Structure and Transfer Learning (2020, methodology): “Although, a recent line of work, e.g. Meta-World and RLBench (Yu et al., 2019; James et al., 2020) aim at studying multi-task learning as well as meta-learning, respective benchmarks mostly exhibit non-parametric hand-designed task variations which makes it ambiguous and not explicit how much…”
SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation (2020, background): “A variety of new environments have been created recently to benchmark reinforcement learning algorithms [15, 34, 35, 36, 37].”
The MAGICAL Benchmark for Robust Imitation (2020, background): “We note that there also exist dedicated IL benchmarks [30, 26], but they are aimed at solving challenging robotics tasks rather than evaluating generalisation directly.”
Online Fast Adaptation and Knowledge Accumulation: a New Approach to Continual Learning (2020, background): “Similarly, it can represent unobserved variables in partially observable Markov decision processes (POMDPs) [32], hidden contexts in hidden-mode Markov decision processes [12], or tasks’ variations in robotics [75, 27, 82].”
SAPIEN: A SimulAted Part-Based Interactive ENvironment (2020, background): “…Dynamic Unity Navigation+, Vision Python, Unity OpenAI Gym MuJoCo [2] Scene-Object Dynamic OpenGL(ﬁxed) Learning, Robotics Python RLBench[22] Scene-Object Dynamic V-REP, PyRep Learning, Vision, Robotics Python, V-REP SAPIEN Scene-Object-Part Dynamic Customizable Learning, Vision,…”
AllenAct: A Framework for Embodied AI Research (2020, methodology): “…[1] and AI Habitat [2]), involving robot-object interaction (such as AI2-THOR [3] and Virtual Home [4]), focused on manipulation (such as RL-Bench [5], Sapien [6], and Meta-world [7]), using advanced physics simulations (such as MuJoCo [8] and ThreeDWorld [9]), and also physical counterparts to…”
3D-VLA: A 3D Vision-Language-Action Generative World Model (2024, positive): “For CALVIN, we compare with MCIL (Lynch & Sermanet, 2020), which is a conditional sequence-to-sequence variational autoencoder. matches the baseline performance in most tasks within the RLBench action prediction, showing its planning capability.”
Orbit: A Unified Simulation Framework for Interactive Robot Learning Environments (2023, methodology): “While various robotic benchmarks have been proposed [1], [4], [5], the right choice of necessary and sufficient tasks to demonstrate “intelligent” behaviors remains an open question.”
BridgeData V2: A Dataset for Robot Learning at Scale (2023, background): “Assembling a large real-world dataset is time-consuming and expensive, so there has also been significant work on developing simulated environments and datasets for robotic manipulation [15, 40, 41, 42] and navigation [43, 44].”
IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks (2019, methodology): “Recent advancements in simulators, such as Robosuite [4], RLBench [2], PyRoboLearn [27], and MetaWorld [3], take a step towards a comprehensive manipulation simulator by offering a variety of manipulation tasks.”
RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots (2024, background): “Some are limited to tabletop manipulation [33, 15, 6, 28], but RoboCasa, along with others [21, 40, 26, 27, 10, 34], support mobile manipulation.”
LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning (2023, neutral): “Robot Learning Benchmarks A variety of robot learning benchmarks have been proposed to address challenges in meta learning (MetaWorld [73]), causality learning (CausalWorld [1]), multi-task learning [27, 35], policy generalization to unseen objects [47, 24], and compositional learning [44].”
VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models (2023, methodology): “We provide an open-sourced implementation of VoxPoser at github.com/huangwl18/VoxPoser based on RLBench [134], as its diversity of tasks and scenes best resembles our real-world setup.”
The ThreeDWorld Transport Challenge: A Visually Guided Task-and-Motion Planning Benchmark Towards Physically Realistic Embodied AI (2021, neutral): “RL-Benchmark [21], Meta-World [51] and CLEVRER [50, 11]).”
Physically Embedded Planning Problems: New Challenges for Reinforcement Learning (2020, background): “…on short-horizon single task challenges (Brockman et al., 2016; Tassa et al., 2018), more recently released benchmark suites (e.g. Fan et al., 2018; James et al., 2020; Yu et al., 2019) focus on multi-task scenarios, where the same embodiment needs to solve a diverse set of challenges or is…”
ManiSkill2: A Unified Benchmark for Generalizable Manipulation Skills (2023, methodology): “2) PlugCharger : a dual peg-in-hole assembly task inspired by RLBench (James et al., 2020), which involves an agent picking up and plugging a charger into a vertical receptacle.”