Codebase for sequentially selecting primitives and their respective goal locations via Deep Reinforcement Learning
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Baselines/DDPGRL
CoRLRealImages
DeterministicParsing
ILRL_Code/GreedyDagger
LSTMParsing
OrganizedCode
ParsingImitation
scripts
README.md

README.md

Learning Neural Parsers with Deterministic Differentiable Imitation Learning

What is this all about?

This repository is code connected to the paper - T. Shankar, N. Rhinehart, K. Muelling, K. M. Kitani, Learning Neural Parsers with Deterministic Differentiable Imitation Learning, submitted to the Conference on Robot Learning (CoRL) 2018.

Where can I read this cool paper?

Click here to view the paper on ArXiv!

I want a TL;DR of what this repository does.

Our paper targets learning a parser to construct hierarchical decompositions of object images, by imitating a decision tree like algorithm. This repository implements the code for the following:

  1. Introduces a framework to learn to decompose spatial tasks into segments by parsing, motivated by the problem of a painting robot covering a large object.
  2. Formulates object decomposition as a parsing problem, inspired by the similarity between parse-trees of objects and structured decisions trees constructed by ID3.
  3. Trains a neural object parser to construct hierarchical object decompositions by imitating a ground-truth expert, in the form of a information gain maximizing ID3 like algorithm.
  4. Introduces a novel hybrid imitation-reinforcement learning approach, by building a deterministic DDPG-style actor-critic variant of AggreVaTeD.

Is that all?

Yes and no. I'm evaluating our novel policy gradient update, DRAG, on OpenAI Gym environments! you can check out https://github.com/tanmayshankar/GymRobotics for more!

Can I use this code to pretend I did some research?

Feel free to use my code, but please remember to cite my paper above (and this repository)!

I have a brilliant idea to make this even cooler!

Awesome! Feel free to mail me at tanmay.shankar@gmail.com with your suggestions, feedback and ideas!