Learn Proportional Derivative Controllable Latent Space from Pixels

Abstract

Recent advances in latent space dynamics model from pixels show promising progress in vision-based model predictive control (MPC). However, executing MPC in real time can be challenging due to its intensive computational cost in each timestep. We propose to introduce additional learning objectives to enforce that the learned latent space is proportional derivative controllable. In execution time, the simple PD-controller can be applied directly to the latent space encoded from pixels, to produce simple and effective control to systems with visual observations. We show that our method outperforms baseline methods to produce robust goal reaching and trajectory tracking in various environments.

References in corpus (14)

• Tutorial on Variational Autoencoders
• DeepMind Control Suite
• Variational Autoencoder for Deep Learning of Images, Labels and Captions
• Learning Latent Dynamics for Planning from Pixels
• Hindsight Experience Replay
• Dream to Control: Learning Behaviors by Latent Imagination
• SOLAR: Deep Structured Representations for Model-Based Reinforcement Learning
• Learning Compositional Koopman Operators for Model-Based Control
• PVEs: Position-Velocity Encoders for Unsupervised Learning of Structured State Representations
• Learning Stability Certificates from Data
• Hamiltonian Generative Networks
• Robust Locally-Linear Controllable Embedding
• Predictive Coding for Locally-Linear Control
• Neural Lyapunov Model Predictive Control: Learning Safe Global Controllers from Sub-optimal Examples