1

Meeting Notes 23.03.2022 – Adna (TRAIN)

Meeting Details

Date : 23 March 2021

Time : 10:00 – 11:00

Location : Webex

Participants: Fotios Lygerakis, Adna Blieck

Agenda

  • Get to know about Adna’s work
  • use of RL algorithms
  • objectives
  • problem formalization

Notes

  • current objective: test different RL algorithms before start working with the robot
  • Current toy environment: Atari Seaquest
  • RL algorithms tested
    • DDQN
    • IQN
    • FQF
  • Learning from human demonstrations
    • get them in a reply buffer and give them high priority over the rest
  • ultimate goal: “how human rates an algorithm”
    • sense of agency
    • trust