SLAM Challenge

Challenge

Video Link

Simultaneous Localization and Mapping (SLAM) is one of the most fundamental capabilities necessary for robots. Due to the ubiquitous availability of images, Visual SLAM (V-SLAM) has become an important component of many autonomous systems. Impressive progress has been made with both geometric-based methods and learning-based methods. However, developing robust and reliable SLAM methods for real-world applications is still a challenging problem. Real-life environments are full of difficult cases such as changing or lack of illumination, dynamic objects, and texture-less scenes. Since data annotation as well as varied enough data collection in real-life data is difficult topic, the TartanAir dataset adopts AirSim, which allows us to collect photo-realistic data in simulation, and gather a much wider range of appearance, sizes, and motion diversity that impossible otherwise. Based on the TartanAir dataset, this challenge provides evaluation tools and testing data for SLAM algorithms with a special focus on aforementioned challenging features. It consists of a monocular track and a stereo track. Each track contains 16 trajectories, which are further divided into easy and hard categories. Participants can attend both tracks or a single track. The results will be evaluated by the average performance on each track.

Submission

You can participate in the competition here:

• Monocular Track

• Stereo Track

Competition Results:

Champion team of Monocular/Stereo Track: MEGVII_SLAM

• Members: Haotian Zhang, Zheng Chai, and Xiao Liu

Runner-up team of Monocular/Stereo Track: Burnt_GPU_Need_Bucks

• Member: Zhixiang Min

Watch online the award ceremony and technical sharing of the winning teams: video link.

Dataset

This challenge is based on the TartanAir dataset. The four most important features of this dataset are:

• Large size diverse realistic data. We collect the data in diverse environments with different styles, covering indoor/outdoor, different weather, different seasons, urban/rural.

• Multimodal ground truth labels. We provide RGB stereo, depth, optical flow, and semantic segmentation images, which facilitates the training and evaluation of various visual SLAM methods

• Diversity of motion patterns. The existing popular datasets such as KITTI and Cityscapes only covers very limited motion patterns, which are mostly moving straight forward plus small left or right turns. This regular motion is too simple to sufficiently test a visual SLAM algorithm. Our dataset covers much more diverse motion combinations in 3D space, which is significantly more difficult than existing datasets.

• Challenging Scenes. We include challenging scenes with difficult lighting conditions, day-night alternating, low illumination, weather effects (rain, snow, wind and fog) and seasonal changes.

Deadline

The deadline for submitting to any of the three challenges is Aug 15th, 23:59 PM (PST).

Prizes

• The prize for the winner: $1500 USD

• The prize for the runner up: $1000 USD

Contact

Wenshan Wang - wenshanw@andrew.cmu.edu

Sebastian Scherer - basti@cmu.edu

Ashish Kapoor - akapoor@microsoft.com