Self-Supervised Visual Planning with Temporal Skip Connections

Abstract

In order to autonomously learn wide repertoires of complex skills, robots must be able to learn from their own autonomously collected data, without human supervision. One learning signal that is always available for autonomously collected data is prediction: if a robot can learn to predict the future, it can use this predictive model to take actions to produce desired outcomes, such as moving an object to a particular location. However, in complex open-world scenarios, designing a representation for prediction is difficult. In this work, we instead aim to enable self-supervised robotic learning through direct video prediction: instead of attempting to design a good representation, we directly predict what the robot will see next, and then use this model to achieve desired goals. A key challenge in video prediction for robotic manipulation is handling complex spatial arrangements such as occlusions. To that end, we introduce a video prediction model that can keep track of objects through occlusion by incorporating temporal skip-connections. Together with a novel planning criterion and action space formulation, we demonstrate that this model substantially outperforms prior work on video prediction-based control. Our results show manipulation of objects not seen during training, handling multiple objects, and pushing objects around obstructions. These results represent a significant advance in the range and complexity of skills that can be performed entirely with self-supervised robotic learning.

accepted at the Conference on Robot Learning (CoRL) 2017

References in corpus (9)

• Generating Videos with Scene Dynamics
• Action-Conditional Video Prediction using Deep Networks in Atari Games
• Deep Predictive Coding Networks for Video Prediction and Unsupervised Learning
• Unsupervised Learning for Physical Interaction through Video Prediction
• Uncertainty-Aware Reinforcement Learning for Collision Avoidance
• Learning to Poke by Poking: Experiential Learning of Intuitive Physics
• Recurrent Environment Simulators
• Parallel Multiscale Autoregressive Density Estimation
• Learning to Fly by Crashing

Cited by in corpus (46)

• Visual Reinforcement Learning with Imagined Goals
• Visual Foresight: Model-Based Deep Reinforcement Learning for Vision-Based Robotic Control
• Stochastic Adversarial Video Prediction
• VideoGPT: Video Generation using VQ-VAE and Transformers
• Adversarial Video Generation on Complex Datasets
• Dream to Control: Learning Behaviors by Latent Imagination
• Train Sparsely, Generate Densely: Memory-efficient Unsupervised Training of High-resolution Temporal GAN
• Grasp2Vec: Learning Object Representations from Self-Supervised Grasping
• A Good Image Generator Is What You Need for High-Resolution Video Synthesis
• Keypoints into the Future: Self-Supervised Correspondence in Model-Based Reinforcement Learning
• Few-Shot Goal Inference for Visuomotor Learning and Planning
• VideoFlow: A Conditional Flow-Based Model for Stochastic Video Generation
• Time-Agnostic Prediction: Predicting Predictable Video Frames
• Mastering Atari with Discrete World Models
• Action and Perception as Divergence Minimization
• Learning Latent Plans from Play
• Learning Robotic Manipulation through Visual Planning and Acting
• InMoDeGAN: Interpretable Motion Decomposition Generative Adversarial Network for Video Generation
• Hierarchical Foresight: Self-Supervised Learning of Long-Horizon Tasks via Visual Subgoal Generation
• RoboNet: Large-Scale Multi-Robot Learning
• Experience-Embedded Visual Foresight
• Time Reversal as Self-Supervision
• Learning and Planning with a Semantic Model
• LaMD: Latent Motion Diffusion for Image-Conditional Video Generation
• ROLL: Visual Self-Supervised Reinforcement Learning with Object Reasoning
• Learning Semantic Embedding Spaces for Slicing Vegetables
• Exploring Spatial-Temporal Multi-Frequency Analysis for High-Fidelity and Temporal-Consistency Video Prediction
• Adaptive Skip Intervals: Temporal Abstraction for Recurrent Dynamical Models
• Reasoning and Generalization in RL: A Tool Use Perspective
• Action-Sufficient State Representation Learning for Control with Structural Constraints
• Improving Sequential Latent Variable Models with Autoregressive Flows
• PriorityCut: Occlusion-guided Regularization for Warp-based Image Animation
• Deep Video Prediction for Time Series Forecasting
• Learning Temporal Dynamics from Cycles in Narrated Video
• Diverse Video Generation using a Gaussian Process Trigger
• Modular Action Concept Grounding in Semantic Video Prediction
• Predicting the Future with Transformational States
• Learning the Reward Function for a Misspecified Model
• Learning to See before Learning to Act: Visual Pre-training for Manipulation
• Dual-MTGAN: Stochastic and Deterministic Motion Transfer for Image-to-Video Synthesis
• Depth by Poking: Learning to Estimate Depth from Self-Supervised Grasping
• Stochastic Dynamics for Video Infilling
• Cross-Identity Motion Transfer for Arbitrary Objects through Pose-Attentive Video Reassembling
• Morphology-Agnostic Visual Robotic Control
• Cubic LSTMs for Video Prediction
• CWY Parametrization: a Solution for Parallelized Optimization of Orthogonal and Stiefel Matrices