Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Motion Generation Using Uncertainty-Driven Foresight Prediction

Hyogo Hiruma, Hiroshi Ito, Tetusya Ogata

TL;DR

The result indicated that the foresight module biased the model to consider future consequences, which lead to embedding uncertainties at the policy of the robot controller, rather than the resultant observation, which is beneficial for implementing adaptive behaviors, which indices derivation of diverse motion during exploration.

Abstract

Uncertainty of environments has long been a difficult characteristic to handle, when performing real-world robot tasks. This is because the uncertainty produces unexpected observations that cannot be covered by manual scripting. Learning based robot controlling methods are a promising approach for generating flexible motions against unknown situations, but still tend to suffer under uncertainty due to its deterministic nature. In order to adaptively perform the target task under such conditions, the robot control model must be able to accurately understand the possible uncertainty, and to exploratively derive the optimal action that minimizes such uncertainty. This paper extended an existing predictive learning based robot control method, which employ foresight prediction using dynamic internal simulation. The foresight module refines the model's hidden states by sampling multiple possible futures and replace with the one that led to the lower future uncertainty. The adaptiveness of the model was evaluated on a door opening task. The door can be opened either by pushing, pulling, or sliding, but robot cannot visually distinguish which way, and is required to adapt on the fly. The results showed that the proposed model adaptively diverged its motion through interaction with the door, whereas conventional methods failed to stably diverge. The models were analyzed on Lyapunov exponents of RNN hidden states which reflect the possible divergence at each time step during task execution. The result indicated that the foresight module biased the model to consider future consequences, which lead to embedding uncertainties at the policy of the robot controller, rather than the resultant observation. This is beneficial for implementing adaptive behaviors, which indices derivation of diverse motion during exploration.

Adaptive Motion Generation Using Uncertainty-Driven Foresight Prediction

TL;DR

The result indicated that the foresight module biased the model to consider future consequences, which lead to embedding uncertainties at the policy of the robot controller, rather than the resultant observation, which is beneficial for implementing adaptive behaviors, which indices derivation of diverse motion during exploration.

Abstract

Uncertainty of environments has long been a difficult characteristic to handle, when performing real-world robot tasks. This is because the uncertainty produces unexpected observations that cannot be covered by manual scripting. Learning based robot controlling methods are a promising approach for generating flexible motions against unknown situations, but still tend to suffer under uncertainty due to its deterministic nature. In order to adaptively perform the target task under such conditions, the robot control model must be able to accurately understand the possible uncertainty, and to exploratively derive the optimal action that minimizes such uncertainty. This paper extended an existing predictive learning based robot control method, which employ foresight prediction using dynamic internal simulation. The foresight module refines the model's hidden states by sampling multiple possible futures and replace with the one that led to the lower future uncertainty. The adaptiveness of the model was evaluated on a door opening task. The door can be opened either by pushing, pulling, or sliding, but robot cannot visually distinguish which way, and is required to adapt on the fly. The results showed that the proposed model adaptively diverged its motion through interaction with the door, whereas conventional methods failed to stably diverge. The models were analyzed on Lyapunov exponents of RNN hidden states which reflect the possible divergence at each time step during task execution. The result indicated that the foresight module biased the model to consider future consequences, which lead to embedding uncertainties at the policy of the robot controller, rather than the resultant observation. This is beneficial for implementing adaptive behaviors, which indices derivation of diverse motion during exploration.
Paper Structure (3 sections, 1 equation, 4 figures)

This paper contains 3 sections, 1 equation, 4 figures.

Table of Contents

  1. INTRODUCTION
  2. METHOD
  3. EXPERIMENT AND RESULTS

Figures (4)

  • Figure 1: Network structure of the proposed model. (A) The model extends conventional RNN models with a foresight module that refines the RNN hidden states through closed-loop predictions. (B) The hidden states are refined by selecting the best noise that led to a foresight with the lowest expected variance.
  • Figure 2: Door opening task environment. The door can be opened either by pushing, pulling, or sliding.
  • Figure 3: Transitioning RNN hidden states per motion type. States are compressed to two dimensions, using principal component analysis.
  • Figure 4: Comparison of per-time-step Lyapunov exponents during motion prediction. Higher values represent that stronger chaotic properties are embedded at that timing.