A Physics-Based Model Prior for Object-Oriented MDPs

By Jonathan Scholz, Martin Levihn, Charles Lee Isbell Jr., David Wingate. 2014

67 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Unsupervised state representation learning with robotic priors: a robustness benchmark (2017, methodology): “The priors we use (as in [1]) are similar to those in [5] because they are physically grounded and aim at building representations of the world that are consistent with physics.”
Structured agents for physical construction (2019, background): “A main direction we explore is object-centric representations of the scene and agent’s actions (Diuk et al., 2008; Scholz et al., 2014), implemented with graph neural networks (Scarselli et al., 2009; Bronstein et al., 2017; Gilmer et al., 2017; Battaglia et al., 2018).”
Schema Networks: Zero-shot Transfer with a Generative Causal Model of Intuitive Physics (2017, positive): “Schema Networks build upon the ideas of the Object-Oriented Markov Decision Process (OO-MDP) introduced by Diuk et al. (2008) (see also Scholz et al. (2014)).”
Reasoning About Physical Interactions with Object-Oriented Prediction and Planning (2018, background): “Object-centric scene decompositions have also been considered as a potential state representation in reinforcement learning (Diuk et al., 2008; Scholz et al., 2014; Devin et al., 2017; Goel et al., 2018; Keramati et al., 2018).”
End-to-End Learnable Histogram Filters (2017, background): “Physics, for example, governs the interaction of any robot and its environment and physics-based priors can substantially improve learning (Scholz et al., 2014; Jonschkowski & Brock, 2015).”
Task-Agnostic Dynamics Priors for Deep Reinforcement Learning (2019, background): “…recent papers have explored the idea of incorporating physics priors into learning dynamics models of environments (Nguyen-Tuong and Peters, 2010; Cutler et al., 2014; Cutler and How, 2015; Scholz et al., 2014; Kansky et al., 2017; Battaglia et al., 2016; Mrowca et al., 2018; Xie et al., 2016).”
Fast Model Identification via Physics Engines for Data-Efficient Policy Search (2017, methodology): “Variants have been applied in robotics [Dogar et al. , 2012; Lynch and Mason, 1996; Scholz et al. , 2014; Zhou et al. , 2016], where using inaccurate models can still allow a policy to steer an RC car [ Abbeel et al. , 2006].”
Unsupervised Video Object Segmentation for Deep Reinforcement Learning (2018, methodology): “The framework was extended with a physics based model to reason about the dynamics of objects [25].”
PVEs: Position-Velocity Encoders for Unsupervised Learning of Structured State Representations (2017, methodology): “Our extension to position-velocity states is inspired by work on physics-based priors in model-based reinforcement learning by Scholz et al. [16], which proposed to learn a physically plausible dynamics model given a position-velocity representation.”
Learning state representations with robotic priors (2015, background): “Scholz et al. [28] follow the same idea of using physics based priors to learn a forward model (or transition function) of the world.”
Learning Actionable Representations from Visual Observations (2018, methodology): “[11] used physics based priors for representation learning and showed that incorporating these improved performance for model-based RL.”
A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms (2019, background): “Bayesian approaches explicitly capture the uncertainty in their predictions by representing the probability distributions over different potential models [84, 337].”
Lessons from the Amazon Picking Challenge: Four Aspects of Building Robotic Systems (2016, background): “In reinforcement learning, adding explicit knowledge about physics makes the learning problem tractable by reducing its dimensionality [25, 44].”
Learning to Manipulate Unknown Objects in Clutter by Reinforcement (2015, methodology): “Also, Scholz et al. (2014) used a Markov Decision Process (MDP) for modeling interactions between objects.”