CASSIA
Unlock more intelligent beauty
| Introduction |
How do you make an agent behave well across multiple tasks? How to make agents more generalized? The "AI Olympiad" series of competitions challenges the participants! The competition invites participants to develop an AI to participate in multiple Olympic sports projects, and work together to promote the research of generalized agents. Diverse and interesting competition scenarios, rich winning awards, and exchange opportunities for top academic conferences, the majority of intelligent body algorithm enthusiasts are welcome to participate in the competition.
Registration for the first phase of the competition, held in collaboration with the World Congress on Artificial Intelligence (WAIC), is now open, so come and join!
>>>>
Background
Figure 1. "Olympic-integrated" environment
The "AI Olympiad" has created a rich and interesting competition environment, including football, running, desktop hockey, sumo wrestling and other projects, and will test the participating agents in different environments, fully investigate the generalization of the agents, and work with the majority of participants to promote the research of universal decision-making agents.
>>>>
Contest Introduction
The AI Olympiad will set up two competition stages.
Phase I
The first phase of the competition will be held from April 20, 2022 to May 26, 2022 as part of the World Congress on Artificial Intelligence 2022 (WAIC 2022). The competition is hosted by the Institute of Automation, Chinese Academy of Sciences, and co-organized by Shanghai Digital Brain Science and Technology Research Institute, Peking University Artificial Intelligence Research Institute, Shanghai Jiao Tong University APEX Data and Knowledge Management Laboratory, Shanghai Boyu Education Technology Co., Ltd. and RLChina Reinforcement Learning Community.
Competition official website:
http://jidiai.cn/waic_2022/
Figure 2. First Stage Race Timeline
The first phase of the competition uses the "Olympic-Integrated" environment as the title (see "Introduction to the Competition Topics" for details). The competition is in partnership with the World Congress on Artificial Intelligence (WAIC), and outstanding teams have the opportunity to be recommended to the conference's SAIL Awards.
The competition committee at this stage is composed of a number of academic experts:
Figure 3. Competition Committee
In the first stage of the competition, there will be 1 first prize team, 3 second prize teams and 6 third prize teams, and each team of the first, second and third prizes will receive 25,000, ¥15,000 and ¥5,000 respectively. Excellent participating teams will also receive awards such as AI blind boxes, medals and certificates, and receive opportunities for internal promotion of front-line scientific research institutions and Internet companies to strengthen learning related positions.
Phase II
The second phase of the competition will be part of the 2022 International Conference on Artificial Intelligence (IJCAI-ECAI 2022), still using the "Olympic-integrated" environment as the competition topic, but the difficulty of the competition questions has increased compared to the first stage, more Olympic items will be added to the competition, and the participants will compete with more international players. The second stage of the competition was held from June 1 to June 30.
Competition official website:
http://jidiai.cn/ijcai_2022/。
Figure 4. Second Stage Race Timeline
The winner of the second stage will receive a cash prize of $1000. Excellent participating teams will also receive awards such as AI blind boxes, medals and certificates, and receive opportunities for internal promotion of front-line scientific research institutions and Internet companies to strengthen learning related positions.
>>>>
Introduction to the competition
Figure 5: Dynamic "Olympic-integrated" environment
The competition uses the "Olympic - Integrated" environment as the title, and the main task is to train the agent to complete running, football, table hockey, sumo and other competitions. Agents have the same perception and motor abilities in each competition. The environment and sample agent code used in the competition have all been open sourced, and participants can modify the sample code to improve the level of the agent.
Contest Rules:
● There are two sides in this game, each side controls an elastic orb intelligent body with the same mass and radius.
Agents can collide with each other or walls, but may lose some speed. The agent itself has energy, and the energy consumed at each step is proportional to the applied driving force and displacement.
● The energy of the agent is restored at a fixed rate at the same time, and if the energy decays to zero, the agent fatigues, resulting in the inability to add force.
● The whole game contains the following sub-games. In running, the agent's goal is to reach the end point as quickly as possible on a randomly selected map. In table hockey, the goal of the agent is to hit the ball into the opponent's goal through collision and prevent its own goal from being breached, and the agent can only move in its own field. In soccer, the goal of the agent is the same as the table hockey project, but the agent is able to move freely on the court. In sumo wrestling, the goal of the agent is to avoid falling out of the ring and to try to knock the opponent out of the ring. The games will run the above sub-games in turn.
● When all sub-games are over, the game ends.
● Rewards: Each party receives a +1 reward after winning a sub-game, and a 0-reward for losing a sub-game.
>>>>
How to register
Visit the "and the first" platform ring page: