Announcing the Habitat-Navigation Challenge 2023

[ykarmesh]

We are excited to announce the Habitat Challenge 2023 for the ObjectNav and ImageNav tasks at the CVPR Embodied AI Workshop.

Habitat.Navigation.Challenge.2023.mp4

ObjectNav focuses on egocentric object/scene recognition and a commonsense understanding of object semantics (where is a bed typically located in a house?). This year we are instantiating ObjectNav on the newly released HM3D-Semantics v0.2 dataset.

ImageNav focuses on visual reasoning and embodied instance disambiguation (is the particular chair I observe the same one depicted by the goal image?). We are adding the ImageNav track for the first time and it is also based on the HM3D-Semantics v0.2 scene dataset.

We introduce several changes in the agent config for easier sim2real transfer. We use the HelloRobot Stretch configuration and support your choice of continuous/waypoint/discrete action spaces. All episodes from both tasks can be navigated without traversing between floors.

Check out the starter code for details:
https://github.com/facebookresearch/habitat-challenge

The public leaderboard will be live on EvalAI on March 25th at:
https://eval.ai/web/challenges/challenge-page/1992/overview

[zhai-create]

How to use the semantic ground-truth image segmentation results in HM3D datasets?
When I use semantic_img = observations["semantic"] to call the ground-truth image segmentation results, I find that the output result is a matrix of all zeros, which is visualized as a dark image. How can I solve this problem?