Prediction and Control with Temporal Segment Models

TL;DR

This paper presents a deep generative modeling approach for learning complex nonlinear system dynamics over temporal segments, enabling stable long-horizon predictions and effective trajectory optimization.

Contribution

It introduces a novel method combining convolutional autoregressive models and variational autoencoders to model trajectory distributions conditioned on past and future actions.

Findings

Stable long-horizon predictions for stochastic systems

Effective modeling of uncertainty and noise effects

Improved sample efficiency in trajectory optimization

Abstract

We introduce a method for learning the dynamics of complex nonlinear systems based on deep generative models over temporal segments of states and actions. Unlike dynamics models that operate over individual discrete timesteps, we learn the distribution over future state trajectories conditioned on past state, past action, and planned future action trajectories, as well as a latent prior over action trajectories. Our approach is based on convolutional autoregressive models and variational autoencoders. It makes stable and accurate predictions over long horizons for complex, stochastic systems, effectively expressing uncertainty and modeling the effects of collisions, sensory noise, and action delays. The learned dynamics model and action prior can be used for end-to-end, fully differentiable trajectory optimization and model-based policy optimization, which we use to evaluate the performance and sample-efficiency of our method.