Visible to the public 2020 CPS ChallengeConflict Detection Enabled

2020 NSF Student CPS Challenge -- Mars edition

Visible to the public 

Submitted by srees on Fri, 09/13/2019 - 12:16pm

2020 CPS Challenge

"SoilScope -- Mars edition"

Virtual competition on CPS-VO: May 15-June 15, 2020

 

Introduction

 

Mars 2020 inspired mission scenario for the 2020 NSF CPS Challenge will be a two week virtual event, emulating an autonomous probe deployment science mission by a rover and drone duo, at the Jezero crater landing site. 

Teams will use the OpenUAV simulation testbed through this CPS-VO group. Here is a demo video of the simulation world. 

Why participate?  

  • Develop autonomous systems, in a fun setting.

  • Engage in agile design iterations.

  • Experiment with complex mission scenarios using powerful cloud-based simulation tools.

  • Repurpose solution to other problems, such as searching for a strategic location to deploy and recover a sensor probe for ecological studies.

 

Qualification round will involve only the drone, and ​Phase I and Phase II events will involve drone as well as rover. 

 

Qualification round goals:

Use the drone's down camera, to search for the probe in the terrain (lower plain), recover it, and deploy it at a prescribed local coordinate of x = 25m, y = -35m.  Example ROS launch file is provided, showcasing a docking logic that resembles our electro-permanent magnet based NSF CPS drone design. A demo of the probe docking and and release logic is shown at https://www.youtube.com/watch?v=gPhtxzkg1yw . A sketch of the docking constraint is shown below. 

Screenshot showing drone in Volcanic Tablelands rock scarp terrain, and the images from down camera. The red dot on the right frame is the probe. 

Here is a description of our soil sampling probe designed for ecological mapping, named 'EarthPod' . We emulate this as a cylinder in simulation. 

 

Rules: 

  1. Solutions have to be autonomous, and written with ROS C++ or Python code with the Gazebo/PX4 environment.

  2. The probe location should not be hardcoded in the autonomy code. Teams have to implement a search pattern on the lower plain of the Volcanic Tablelands rock scarp terrain model, to detect the probe.

  3. One minute video of each task finished needs to be submitted, it should show at least three trials.

  4. Code should be maintained on a GitHub repo for persistence and verification.

  5. For qualifying, use of deep neural networks is not allowed. Use of OpenCV is encouraged. Avoid computationally intensive algorithms, since these are harder to deploy on constrained space-grade hardware. 

Phase II objective: 

  1. Use the drone to pick up the sample probe on the Volcanic Tablelands terrain, and deploy it at the location marked by the 'washer' out in the Jezero crater terrain. 
  2. Rendezvous with the traveling rover, and land on its rear rack. 

Rules: 

  1. Do not hardcode the elevation (z coordinate) information for probe, drop location, and rover location.  Instead, use a combination of distance sensors and cameras for safe terrain-relative navigation. 
  2. Assume an error of 10m in the location information for probe and rover. 
  3. Deep neural networks are allowed for detection of the rover, though you are encouraged to explore use of fiducials.   

Notes:

  • mavros rover topic namespace = uav0, and drone topic namespace = uav1
  • rover pose can be subscribed to within drone code, large errors (~50m) may exist due to wheel slippage.
  • to start rover motion (x = 0 m/s, y = 1 m/s in global frame), switch rover to offboard and arm using QGroundControl.
  • RTAB-Map is a good resource for mapping with onboard depth camera. 

Scoring: 

We will test your code with different initial conditions, and observe success rate, mission time, efficiency (sum of squares of rotor commands), and probe deployment accuracy. 

Supplementary:

Welcome webinar, background, rules -- recording, May 18, 2020

 

Software: The project is hosted using the OpenUAV Cloud Testbed. Teams should plan to use the Robot Operating System (ROS) in a UNIX environment. Development can be done exclusively with Python, though C++ could be used as well. More information is available in the competition simulation environment (register to access).  

Anand, Harish, Stephen A. Rees, Zhiang Chen, Ashwin Jose Poruthukaran, Sarah Bearman, Lakshmi Gana Prasad Antervedi, and Jnaneshwar Das. "OpenUAV cloud testbed: a collaborative design studio for field robotics." In 2021 IEEE 17th International Conference on Automation Science and Engineering (CASE), pp. 724-731. IEEE, 2021.


Who owns the code produced by participants ? 

You do, though by design, it will have to be open source under the BSD, Apache 2, or equivalent license. A core goal for this competition is to push the limits on reproducibility of research code.   

 

Interested in participating?

Potential teams are encouraged to join the project (See the "Join Us" balloon in the top-right corner of the header graphic above).

 

2020 SoilScope Mars edition winners!

Planet Porters: Srikar Siddarth and Yogesh, National Institute of Technology, Karnataka, India

The organizers were impressed by how quickly Planet Porters solved the qualifying round with very little prior knowledge of the domain or the system. Their sensing modality (sonar array), and visual-servoing methodology seemed robust. Overall, they demonstrated success for all stages.

Runners-up:

  • Sparky UAV: Muhammad Iqbal, Pratibha Laxmi Darla, Sammeet Thosar and Sampriti Neog
  • Drone Trekkers: Swastik Nandan, Darwin Mick and Rehan Guha

 

Participating Teams:

  Team name Participants Phases participated Phases completed (YouTube video links to solutions)
1. Drone Trekkers Swastik Nandan, Darwin Mick and Rehan Guha Qualifying, Phase 1, Phase 2 Qualifying, Phase 1a, Phase 1b, Phase 2a, Phase 2b
2. Sparky UAV Muhammad Iqbal, Pratibha Laxmi Darla, Sammeet Thosar and Sampriti Neog Qualifying, Phase 1, Phase 2 Qualifying, Phase 1, Phase 2
3. Planet Porters Srikar Siddarth and Yogesh Qualifying, Phase 1, Phase 2 Qualifying, Phase 1, Phase 2a, Phase 2b, Phase 2c 
4. Penn AiR Dragonfly Damian Owerko and Joshua Chen Qualifying, Phase 1 Qualifying
5. Phoenix Abhijith Thottumadayil Jagadeesh and Sreekesh Kattil Chitezhath  Phase 2  Phase 2

Phase II videos:

Phase I videos:

  1. Muhammad Iqbal, Pratibha Laxmi Darla, Sammeet Thosar and Sampriti Neog, Sparky UAV, Arizona State University, USA,  (May 24)  https://www.youtube.com/watch?v=UhxmDlgXcgc
  2. Swastik Nandan and Darwin Mick, Drone Trekkers, Arizona State University, USA, 
    1. (May 24) https://youtu.be/DECznNx3H0Y
    2. (May 26) https://youtu.be/VhFv8HYfAVs
  3. Srikar Siddarth and Yogesh, PlanetPorters, National Institute of Technology, Karnataka, India, (May 25) https://youtu.be/7KBGRbD9ktU

Qualifying team videos: 

  1. Owerko et al., Penn AiR Dragonfly, University of Pennsylvania, USA, https://www.youtube.com/watch?v=s0s5eaYW9tQ
  2. Swastik Nandan and Darwin Mick, Drone Trekkers, Arizona State University, USA, https://www.youtube.com/watch?v=n1WgBF8lem8
  3. Muhammad Iqbal, Pratibha Laxmi Darla, Sammeet Thosar and Sampriti Neog, Sparky UAV, Arizona State University, USA, https://www.youtube.com/watch?v=Bk0NsAASw2s
  4. Srikar Siddarth, National Institute of Technology, Karnataka, India, https://www.youtube.com/watch?v=PagFUjz2GYM  

 

Organizing committee:

Dr. Jnaneshwar Das, Arizona State University

Harish Anand, Arizona State University

 

Technical committee:

Dr. Heather Throop, Arizona State University  

Dr. Hannah Kerner, University of Maryland  

 

If you have questions, please contact Jnaneshwar Das 

 

 

* See Previous CPS Challenges:  https://cps-vo.org/group/arcompetitions