People
- Ludovic Hofer (Rhoban FC)
- Maike Paetzel (Bit-Bots)
- … (other contributors are welcome)
Description
This projects aim at developing three different tools allowing to create large datasets of games usable by the whole league:
- A visualizer to display augmented video streams during and after a match
- Add information from the robot based on the protocol proposed by the NUBots
- Handle multiple cameras
- Allow online streaming
- A labeling tool to annotate games either semi-automatically or automatically
- Humans can provide/confirm position of objects during the game (ball, robots, referees)
- Using tracking algorithms to reduce the tagging burden
- Multiple points of view allow to retrieve 3d position for different objects
- An analyzer to extract high-level metrics from the labeled data.
- Based on information from the GameController and the labeling tool
- Extract high level metrics such as:
- robot peak speed
- Localization accuracy
- Number of pick-ups
Motivations
- Provide more data based on games played during the tournament
- Make games more understandable and engaging (display robot intentions)
- For the public
- For other teams watching the games
- Allow measurement of the in-game performance of robots
- Make tracking of performance along the years easier
- Ensure the event-triggered roadmap is based on reliable information
Status
A first proof of concept has been created and used based on RoboCup 2019 data:
While it already has functional elements it requires drastical modifications:
- It does only support a custom protocol from Rhoban yet
- Performance should strongly be improved to allow
- Ergonomy is basic
- Some of the dependencies are outdated
- Documentation and hands-on tutorial should be made available