The DodgeDrone challenge revisits the popular dodgeball game in the context of autonomous drones. Specifically, participants will have to code navigation policies to fly drones between waypoints while avoiding dynamic obstacles. Drones are fast but fragile systems: as soon as something hits them, they will crash! Since objects will move towards the drone with different speeds and accelerations, smart algorithms are required to avoid them!
The competition consists of two challenges: (i) navigation in a static environment, and (ii) navigation in a dynamic environment. The navigation policy can only rely on on-board perception (dense depth, agent and goal location). These two modalitites will help participants to concentrate on different aspects of the navigation algorithm. Two environments are used for the competition: a simple and irrealistic one (which you should reserve for training and development); and a testing one consising of a photorealistic forest, where drones have to avoid the vegetation, as well as the rocks and birds that will obstruct their path.
You can develop your algorithm and submit solutions according to the instructions provided at the following link: https://github.com/uzh-rpg/flightmare/tree/ddc_challenge.
This challenge provides an unique opportunity to push the current state of the art of agile navigation in dynamic environments. Existing navigation algorithms often struggle in these conditions, since they lack the ability to condition actions to future objects' locations. Competing in the challenge will deepen your expertise in computer vision and control, and boost your research.
In case that’s not enough: the winner of the completition will win a Skydio2! Thank you very much to Skydio for sponsoring this award!
We will evaluate the submitted policies on the following days: April 10th, May 10th, and June the 1st (three days before the workshop). While the first two evaluations are only for you to double check your submission, the last is the one that really counts. You can also only submit on that if you wish. Indeed, the algorithm with highest score in this last evaluation is the one that will win the Skydio2! Note that we will use the score in the forest with dynamic environments to rank the submissions!
We will publish results on this page. Feel free to follow antoniloq on Twitter for updates. Start as soon as possible! Navigation in dynamic environments is a complex problem!
- Call for submissions opens: 12th March 2021
- Submission for first trial: 10th April 2021 (Optional)
- Submission for second trial: 10th May 2021 (Optional)
- Final Submission: 1st June 2021 (Compulsory!)
- Winner announced: June the 4th, during workshop