Unsupervised Discovery of Objects Physical Properties Through Maximum Entropy Reinforcement Learning

Understanding the environment is crucial for autonomous robots to perform navigation and manipulation tasks. Never-seen-before objects may have complex appearances and dynamics, where only physical interactions can help to identify visually hidden properties like mass or friction. In this work we propose a baseline for the newly defined problem of using physical interactions to discover unknown properties of objects, without prior knowledge of them or any supervision. The agent first uses intrinsically motivated unsupervised reinforcement learning to learn how to interact with objects, so as to get observations with a level of information which eases the physical properties estimation. A self-supervised predictive task is then set up while following the learned behaviour to extract a latent representation of the physical properties of an object. When applied to a simulated mobile robot in presence of varying objects, the proposed baseline identifies and differentiates categorical properties, e.g. shape, and quantifies continuous properties, e.g. mass and friction, with excellent correlations to their true values even from noisy observations. It achieves significantly better results than simple interactions of a policy that performs poor exploration. This work provides an implementation of a functional, object-oriented action-perception cycle for embodied robotic agents.

• Publication year

  2025

• Venue

  IEEE Robotics and Automation Letters

• Publication date

  2025-04-01

• Fields of study

  Computer Science, Engineering

• Identifiers
  DOI 10.1109/LRA.2025.3545382
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Bi-path Combination YOLO for Real-time Few-shot Object Detection

  2022cited by this paper
• Real-time Mapping of Physical Scene Properties with an Autonomous Robot Experimenter

  2022cited by this paper
• Predicting Physical Object Properties from Video

  2022cited by this paper
• Behavior From the Void: Unsupervised Active Pre-Training

  2021cited by this paper
• Task-Agnostic Exploration via Policy Gradient of a Non-Parametric State Entropy Estimate

  2021cited by this paper
• Physical Representation Learning and Parameter Identification from Video Using Differentiable Physics

  2021cited by this paper
• URLB: Unsupervised Reinforcement Learning Benchmark

  2021influential reference
• SSFNET-VOS: Semantic segmentation and fusion network for video object segmentation

  2020cited by this paper
• Estimating Mass Distribution of Articulated Objects through Physical Interaction

  2019cited by this paper
• Embodied Amodal Recognition: Learning to Move to Perceive Objects

  2019cited by this paper
• DensePhysNet: Learning Dense Physical Object Representations via Multi-step Dynamic Interactions

  2019cited by this paper
• Unsupervised Learning of Latent Physical Properties Using Perception-Prediction Networks

  2018cited by this paper
• PCN: Point Completion Network

  2018cited by this paper
• Provably Efficient Maximum Entropy Exploration

  2018cited by this paper
• Learning to See Physics via Visual De-animation

  2017cited by this paper
• FiLM: Visual Reasoning with a General Conditioning Layer

  2017cited by this paper
• Active perception model extracting object features from unlabeled data

  2017cited by this paper
• Visual Interaction Networks: Learning a Physics Simulator from Video

  2017cited by this paper
• Learning to Perform Physics Experiments via Deep Reinforcement Learning

  2016cited by this paper
• The Curious Robot: Learning Visual Representations via Physical Interactions

  2016cited by this paper
• PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

  2016cited by this paper
• Learning to Poke by Poking: Experiential Learning of Intuitive Physics

  2016cited by this paper
• Physics 101: Learning Physical Object Properties from Unlabeled Videos

  2016cited by this paper
• Galileo: Perceiving Physical Object Properties by Integrating a Physics Engine with Deep Learning

  2015cited by this paper
• On the Properties of Neural Machine Translation: Encoder–Decoder Approaches

  2014cited by this paper
• Differential entropy estimation by particles

  2011cited by this paper
• Inertia Parameters Estimation of Planar Object on Robot Pushing Operation

  2009cited by this paper
• What is Intrinsic Motivation? A Typology of Computational Approaches

  2007cited by this paper
• Intrinsically Motivated Reinforcement Learning

  2004cited by this paper
• Infants' Physical World

  2004cited by this paper

• No citing papers are available for this paper.