Call for Participation: IROS 2022 Safe Robot Learning (SRL) Competition
SiQi Zhou
Dear all,
We would like to announce that our IROS 2022 Safe Robot Learning (SRL) Competition is now open for submissions. With this competition, our goal is to bring together researchers from different communities to (1) solicit novel and data-efficient robot learning algorithms, (2) establish a common forum to compare control and reinforcement learning approaches for safe robot decision-making, and (3) identify the shortcomings or bottlenecks of the state-of-the-art algorithms with respect to real-world deployment.
We recently released the simulation code for the competition. The submitted algorithms will be tested in our flying arena in Toronto. The competition results will be presented virtually during IROS on October 24-25. A detailed description of the competition including important deadlines is included below.
Best wishes,
SRL Competition Organizing Team
————————————
Code Repository (with Instructions and Details of the Competition Setup):
https://github.com/utiasDSL/safe-control-gym/tree/beta-iros-competition
A Video Illustrating the Evaluation Environment in Simulation and in Real Experiments:
https://www.youtube.com/watch?v=PwphA_jsNKw
Competition Webpage:
https://www.dynsyslab.org/iros-2022-safe-robot-learning-competition/
Competition Description:
The task is to design a controller/planner that enables a quadrotor (Crazyflie 2.x) to safely fly through a set of gates and reach a predefined target despite uncertainties in the robot dynamics (e.g., mass and inertia) and the environment (e.g., wind and position of the gates). The algorithms will be evaluated regarding their safety (e.g., no collisions) and performance (e.g., time to target). We encourage participants to explore both control and reinforcement learning approaches (e.g., robust, adaptive, predictive, learning-based and optimal control, and model-based/model-free reinforcement learning). The controller/planner has access to the position and attitude measurements provided by a motion capture system and the noisy pose of the closest next gate. The controller can send position, velocity, acceleration and heading references to an onboard position controller.
Competition Schedule:
August 29: Initial release of competition tasks in simulation
October 12: Deadline for code submission tested in simulation & selection of finalists
October 12 – October 21: Hardware evaluation tests with finalist teams (remotely on the testbed in Toronto)
October 22: Final code submission deadline
October 24, 25: Demonstration day & announcement of winners