Sim2Real Predictivity: Does Evaluation in Simulation Predict Real-World Performance?
arXiv:1912.06321 · doi:10.1109/LRA.2020.3013848
Abstract
Does progress in simulation translate to progress on robots? If one method outperforms another in simulation, how likely is that trend to hold in reality on a robot? We examine this question for embodied PointGoal navigation, developing engineering tools and a research paradigm for evaluating a simulator by its sim2real predictivity. First, we develop Habitat-PyRobot Bridge (HaPy), a library for seamless execution of identical code on simulated agents and robots, transferring simulation-trained agents to a LoCoBot platform with a one-line code change. Second, we investigate the sim2real predictivity of Habitat-Sim for PointGoal navigation. We 3D-scan a physical lab space to create a virtualized replica, and run parallel tests of 9 different models in reality and simulation. We present a new metric called Sim-vs-Real Correlation Coefficient (SRCC) to quantify predictivity. We find that SRCC for Habitat as used for the CVPR19 challenge is low (0.18 for the success metric), suggesting that performance differences in this simulator-based challenge do not persist after physical deployment. This gap is largely due to AI agents learning to exploit simulator imperfections, abusing collision dynamics to 'slide' along walls, leading to shortcuts through otherwise non-navigable space. Naturally, such exploits do not work in the real world. Our experiments show that it is possible to tune simulation parameters to improve sim2real predictivity (e.g. improving from 0.18 to 0.844), increasing confidence that in-simulation comparisons will translate to deployed systems in reality.
References in corpus (2)
Cited by in corpus (27)
- Bi-directional Domain Adaptation for Sim2Real Transfer of Embodied Navigation Agents
- Embodied Visual Navigation with Automatic Curriculum Learning in Real Environments
- A Survey of Embodied AI: From Simulators to Research Tasks
- VAT-Mart: Learning Visual Action Trajectory Proposals for Manipulating 3D ARTiculated Objects
- Habitat 2.0: Training Home Assistants to Rearrange their Habitat
- Focus on Impact: Indoor Exploration with Intrinsic Motivation
- Actionable Models: Unsupervised Offline Reinforcement Learning of Robotic Skills
- From Seeing to Moving: A Survey on Learning for Visual Indoor Navigation (VIN)
- Analysis of Randomization Effects on Sim2Real Transfer in Reinforcement Learning for Robotic Manipulation Tasks
- OpenRooms: An End-to-End Open Framework for Photorealistic Indoor Scene Datasets
- Out of the Box: Embodied Navigation in the Real World
- Hardware-in-the-Loop Simulation for Evaluating Communication Impacts on the Wireless-Network-Controlled Robots
- Mindstorms in Natural Language-Based Societies of Mind
- Bridging the Domain Gap for Flight-Ready Spaceborne Vision
- AI Simulation by Digital Twins: Systematic Survey, Reference Framework, and Mapping to a Standardized Architecture
- Simultaneous Human-robot Matching and Routing for Multi-robot Tour Guiding under Time Uncertainty
- Visual Semantic Navigation with Real Robots
- The Audio-Visual BatVision Dataset for Research on Sight and Sound
- Surgical Vision World Model
- Spot the Difference: A Novel Task for Embodied Agents in Changing Environments
- Scalable and low-cost remote lab platforms: Teaching industrial robotics using open-source tools and understanding its social implications
- SIM2REALVIZ: Visualizing the Sim2Real Gap in Robot Ego-Pose Estimation
- RLOps: Development Life-cycle of Reinforcement Learning Aided Open RAN
- VHAKG: A Multi-modal Knowledge Graph Based on Synchronized Multi-view Videos of Daily Activities
- Accelerating the Development of Multimodal, Integrative-AI Systems with Platform for Situated Intelligence
- Curriculum-based Sensing Reduction in Simulation to Real-World Transfer for In-hand Manipulation
- Reinforcement Learning of Dolly-In Filming Using a Ground-Based Robot