Register now After registration you will be able to apply for this opportunity online.
Minimal: Learning-based design optimization for robots over challenging terrain
Minimal is a mostly 3D-printed, highly reconfigurable robot. Using state-of-the-art reinforcement learning, we will explore novel and highly advanced hardware design possibilities that will be coupled with design optimization through learning. This will enable the next generation of robots to be a lot faster, stronger and agile.
Keywords: Robot design optimization, deep reinforcement learning, optimization, legged robots
Traditionally, legged robot design was focusing on creating a simple design with as little mass leg as possible to make the model-based control with reduced-order models of these robots easier. Nowadays with the emergence of reinforcement learning, legged robot control is very robust and has the capabilities of easily exploiting the robot's hardware for fast and effective locomotion. Therefore, leveraging novel and complex design approaches is feasible. Particularly, we are interested in novel design approaches for legged robots that highly increase the efficiency, speed, strength and agility of the machines. Possible new design approaches could be Series Elastic Actuators, Parallel Elastic Actuators, Series-Parallel Elastic Actuators, soft materials, multi linkage design, parallel linkage design, actuated ankles and many more. The student will come up with different robot concepts, analyse and optimize them with state-of-the-art tools (e.g. learning based approaches) with the possiblity to implement promising concepts in hardware.
Traditionally, legged robot design was focusing on creating a simple design with as little mass leg as possible to make the model-based control with reduced-order models of these robots easier. Nowadays with the emergence of reinforcement learning, legged robot control is very robust and has the capabilities of easily exploiting the robot's hardware for fast and effective locomotion. Therefore, leveraging novel and complex design approaches is feasible. Particularly, we are interested in novel design approaches for legged robots that highly increase the efficiency, speed, strength and agility of the machines. Possible new design approaches could be Series Elastic Actuators, Parallel Elastic Actuators, Series-Parallel Elastic Actuators, soft materials, multi linkage design, parallel linkage design, actuated ankles and many more. The student will come up with different robot concepts, analyse and optimize them with state-of-the-art tools (e.g. learning based approaches) with the possiblity to implement promising concepts in hardware.
Possible topics:
- Fast running
- Agile rough terrain and stair climbing
- Efficient and natural locomotion
- Hybrid quadrupedal-bipedal mode
- Reinforcement Learning
- Hardware Design
Do you have your own idea in mind? We are happy to discuss and support different topics with your own research questions!
Possible topics:
- Fast running - Agile rough terrain and stair climbing - Efficient and natural locomotion - Hybrid quadrupedal-bipedal mode - Reinforcement Learning - Hardware Design
Do you have your own idea in mind? We are happy to discuss and support different topics with your own research questions!
Your application should include a brief motivational statement, your transcript of records and your CV.
Filip Bjelonic (fbjelonic@ethz.ch)
Fabian Tischhauser (tfabian@mavt.ethz.ch)
Your application should include a brief motivational statement, your transcript of records and your CV. Filip Bjelonic (fbjelonic@ethz.ch) Fabian Tischhauser (tfabian@mavt.ethz.ch)
