πŸ““ New Multitask Hierarchical Baseline for RL Task and πŸ† additional $1,500 prizes

:wave: Hello everyone,

Here are a few updates for NeurIPS 2022: IGLU Challenge.

:notebook_with_decorative_cover: New Baseline For RL Task

The Multitask Hierarchical Baseline (MHB) agent fully solves the IGLU RL task: it predicts which blocks needs to be placed from instruction, and then, it executes several actions in the environment that lead to placements in response to that instruction.

It is comprised of three modules: the NLP module that predicts blocks coordinates and ids given text, the Heuristic module (pure python code) that iterates over the predicted blocks (where a heuristic defines order), and the RL module that executes the atomic task of one block placement or removal. The RL agent operates on visual input, inventory, compass observation, and a target block.

Read about the methods and visualisation over here. Install and run the new baseline to make your submission.


:trophy: Additional $1,500 in Prizes

The organisers have added an additional $1,500 to the existing $15,000 prize pool! :dancer:
The Research Prize Pool is now for $3,500. To participate in this prize, please include details on interesting research-relevant results in the README for your submission. This will be handed out at the organizers’ discretion to submissions that we think made a particularly interesting or valuable research contribution.

Updated Prizes for Third Place

The prizes for third place on the final leaderboard for RL Task and NLP Task is now $1,000 instead of $500.

The new prize pool is divided as follows:

NLP Task

  • 1st place: $4,000 USD
  • 2nd place: $1,500 USD
  • 3st place: $1,000 USD

RL Task

  • 1st place: $4,000 USD
  • 2nd place: $1,500 USD
  • 3st place: $1,000 USD

Research prizes: $3,500 USD

1 Like