Fra Robin

Revisjon per 29. aug 2014 kl. 08:11 av Snorreah (Diskusjon | bidrag)
(diff) ← Eldre versjon | Nåværende versjon (diff) | Nyere versjon → (diff)
Gå til: navigasjon, søk

Description of master project

Robotics: Nao as a platform for evolutionary robotics.

Evolutionary robotics is a method of automatic design and tuning of robotic behavior and design. It has the potential of making the design process easier for a human designer and also to propose original and competitive solutions. Common challenges of evolutionary robotics experiments include: - The real robot getting stuck or falling over during evaluation - Wear and tear of the real robot - Accuracy of the simulation model Nao is a humanoid robot commonly used by universities for research and education, and is the standard platform for the Robocup. With seemingly robust capabilities, including the ability to recover from a fall, the robot seems promising for experiments with evolutionary learning methods. We would like to perform initial investigations of Nao's suitability as an evolutionary robotics platform.

The project consists of developing a setup for evolutionary learning which should include the following properties: - A learning module with possibility for testing various algorithms - Possibility for evaluating a solution both on the real robot and in a simulator - Integrated data collection from the experiments, both for the learning loop as well as for reporting. This would include motion capture data from testing on the real robot.

Milestones of the project include: - Investigation into previous research on evolutionary robotics, in particular for real life platforms, humanoid robots, and the Nao platform in particular. Writing essay. - Investigating the capabilities of the currently available software package(s) and its suitability for simulation, interfacing and integration in an evolutionary learning loop. - Implementing an integrated evolutionary learning platform based on existing and/or self-developed components. The interfaces and behavior of the platform should be documented for future users. - Testing the platform by carrying out experiments on locomotion learning, such as walking, crawling, or similar. This involves choosing a suitable control system and a suitable learning algorithm.

Personlige verktøy