Backward Learning for Goal-Conditioned Policies

Marc Höftmann, Jan Robine, Stefan Harmeling
12/8/2023
cs.LGcs.AI

No official code found. Best implementation: Hauf3n/Backward-Learning-for-Goal-Conditioned-Policies (1 star, MIT, last commit 2024-06-06).

Abstract

Can we learn policies in reinforcement learning without rewards? Can we learn a policy just by trying to reach a goal state? We answer these questions positively by proposing a multi-step procedure that first learns a world model that goes backward in time, secondly generates goal-reaching backward trajectories, thirdly improves those sequences using shortest path finding algorithms, and finally trains a neural network policy by imitation learning. We evaluate our method on a deterministic maze environment where the observations are $64\times 64$ pixel bird's eye images and can show that it consistently reaches several goals.

View on arXivView PDF

Code Implementations(1)

Hauf3n/Backward-Learning-for-Goal-Conditioned-PoliciesCommunity100%
MIT

Code for Backward Learning for Goal-Conditioned Policies

10PythonApr 11, 20242 years agoMIT

Similar Papers

Mitigating Off-Policy Bias in Actor-Critic Methods with One-Step Q-learning: A Novel Correction Approach

86% match1 repo92022

Tackling GNARLy Problems: Graph Neural Algorithmic Reasoning Reimagined through Reinforcement Learning

85% match2025

Understanding and Controlling a Maze-Solving Policy Network

85% match2023

An Agency-Transferring Model-Free Policy Enhancement Technique

85% match1 repo2026

ChronoSRL: Temporal Geometry for Self-Supervised Reinforcement Learning

85% match2026

Neural-Network-Driven Reward Prediction as a Heuristic: Advancing Q-Learning for Mobile Robot Path Planning

85% match2024

Cite this paper

@article{hftmann2023backward,
  title  = {Backward Learning for Goal-Conditioned Policies},
  author = {Marc Höftmann and Jan Robine and Stefan Harmeling},
  year   = {2023},
  eprint = {2312.05044},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2312.05044v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2312.05044.svg)](https://codeofpaper.com/paper/2312.05044)

Discussion