Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Stimulate the Potential of Robots via Competition

Kangyao Huang, Di Guo, Xinyu Zhang, Xiangyang Ji, Huaping Liu

TL;DR

A competitive learning framework which is able to help individual robot to acquire knowledge from the competition, fully stimulating its dynamics potential in the race by introducing the competition information among competitors as the additional auxiliary signal to learn advantaged actions.

Abstract

It is common for us to feel pressure in a competition environment, which arises from the desire to obtain success comparing with other individuals or opponents. Although we might get anxious under the pressure, it could also be a drive for us to stimulate our potentials to the best in order to keep up with others. Inspired by this, we propose a competitive learning framework which is able to help individual robot to acquire knowledge from the competition, fully stimulating its dynamics potential in the race. Specifically, the competition information among competitors is introduced as the additional auxiliary signal to learn advantaged actions. We further build a Multiagent-Race environment, and extensive experiments are conducted, demonstrating that robots trained in competitive environments outperform ones that are trained with SoTA algorithms in single robot environment.

Stimulate the Potential of Robots via Competition

TL;DR

A competitive learning framework which is able to help individual robot to acquire knowledge from the competition, fully stimulating its dynamics potential in the race by introducing the competition information among competitors as the additional auxiliary signal to learn advantaged actions.

Abstract

It is common for us to feel pressure in a competition environment, which arises from the desire to obtain success comparing with other individuals or opponents. Although we might get anxious under the pressure, it could also be a drive for us to stimulate our potentials to the best in order to keep up with others. Inspired by this, we propose a competitive learning framework which is able to help individual robot to acquire knowledge from the competition, fully stimulating its dynamics potential in the race. Specifically, the competition information among competitors is introduced as the additional auxiliary signal to learn advantaged actions. We further build a Multiagent-Race environment, and extensive experiments are conducted, demonstrating that robots trained in competitive environments outperform ones that are trained with SoTA algorithms in single robot environment.
Paper Structure (18 sections, 8 equations, 6 figures, 1 table)

This paper contains 18 sections, 8 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Competition in Multi-agent Task
  4. Learning from Competitive and Adversarial Data
  5. Problem Formulation
  6. Methodology
  7. Framework
  8. Policy Training $\&$ Experience Sharing
  9. Robot Observation Construction
  10. Evaluation
  11. Emperical Results
  12. Environment
  13. Baselines
  14. Experimental Details
  15. Role of Competition
  16. ...and 3 more sections

Figures (6)

  • Figure 1: Episode reward comparison between competitive and non-competitive Walker2d environment. The performance of Walker2ds trained with competition can reach 120$\%$ of the baseline.
  • Figure 2: Framework for learning knowledge from comparative information. Where $a$ denotes actions, $s$ represents the proprioceptive state, and $o$ is the competitive observation.
  • Figure 3: The proposed Self-Interest Competition Environments Multiagent-Race. Three walker2d robots racing is illustrated as an example.
  • Figure 4: Performance of different settings on 3-agent environments. $S$A: Yellow. $3$A-Sh-Decent: Blue. $3$A-Sh-Decent-Comp: Red. $3$A-Sp-Decent-Comp: Green. $3$A-Sh-Cent-Comp: Brown.
  • Figure 5: Comparison between $3$A-Sh-Decent (Non: non-competitive information as inputs), $3$A-Sh-Decent-Noi (Noi: noise information as inputs), and $3$A-Sh-Decent-Comp (Comp: competitive information as inputs).
  • ...and 1 more figures