Sim2Real Predictivity: Does Evaluation in Simulation Predict Real-World Performance?

By Abhishek Kadian, Joanne Truong, Aaron Gokaslan, Alexander Clegg, Erik Wijmans, Stefan Lee, Manolis Savva, Sonia Chernova, Dhruv Batra. 2019

263 citations recorded by the source corpus.

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 M. Savva et al., 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 SRCC$_{\text{Succ}}$ from 0.18 to 0.844) – increasing confidence that in-simulation comparisons will translate to deployed systems in reality.

Subfields: Control and Systems Engineering, Management Science and Operations Research, Ocean 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 Control and Systems Engineering

In Management Science and Operations Research

In Ocean Engineering

In Evaluation & Benchmarks

In Robotics & Control

Citations in the corpus

Habitat 2.0: Training Home Assistants to Rearrange their Habitat (2021, methodology): “In addition, we make several optimizations: (1) We employ a navigation mesh to move the robot base kinematically (which has been show to transfer well to real the world [64]) rather than simulating wheel-ground contact.”
A Survey of Embodied AI: From Simulators to Research Tasks (2021, background): “However, Sim2Real [27]–[29] and robotics in the physical world will not be covered in this paper.”
From Seeing to Moving: A Survey on Learning for Visual Indoor Navigation (VIN) (2020, result): “To this end, many works study the simulation to real transfer in the visual navigation domain [Bharadhwaj et al., 2019], while in [Kadian et al., 2019], the authors suggest to evaluate simulators in terms of how likely the performance improvement achieved on them can hold in reality, rather than…”
Evaluating Real-World Robot Manipulation Policies in Simulation (2024, unknown): “…pipelines that fail to linearly recover the real results despite recovering the correct ranking, and it is overly sensitive to minor noise in evaluations when different policies perform similarly in the real world. evaluations, we want the same comparison to hold in our simulated evaluation.”
Tool Learning with Foundation Models (2023, unknown): “To narrow the gap between sim-to-real transfer (Kadian et al., 2020) and enhance agent performance, it is essential to incorporate embodied tools within simulated environments.”
ViNT: A Foundation Model for Visual Navigation (2023, background): “I2O and related efforts [15, 16] show impressive transfer from simulation to real-world environments, but we emphasize that our aim is orthogonal to the specific choice of algorithm: we focus on learning a capable navigation policy that can be efficiently adapted to solve different downstream tasks.”
ProcTHOR: Large-Scale Embodied AI Using Procedural Generation (2022, methodology): “Following common practice [27, 54], we use stochastic actuation to better simulate noise in the real world.”
OpenRooms: An Open Framework for Photorealistic Indoor Scene Datasets (2020, background): “Since we create the scene from scans, correspondence is available to real scenes, which may be useful for sim-to-real transfer studies [27].”
NoMaD: Goal Masked Diffusion Policies for Navigation and Exploration (2023, background): “Several prior works have proposed learning-based approaches for robotic exploration by leveraging privileged information in simulation [13–16] or learn directly from real-world experience [17].”
Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (2024, background): “Some works reported poor transfer due to visual domain differences (82), while others found success through parameter tuning (85), abstraction of dynamics (86), or employing only depth images rather than RGB-D (83, 87).”