Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
RL Chassis Balancing on HEAP
The goal of this project is to learn a control policy using reinforcement learning (RL) that adjusts the height and the orientation of the cabin of a 12t walking excavator. The policy will be trained in simulation and transferred to the machine for deployment (sim-to-real).
Keywords: reinforcement learning, deep learning, control, excavation, hydraulic, construction robotics
To facilitate the operation of walking excavators a chassis balancing controller has been developed that optimizes the wheel contact force distribution given high-level inputs such as cabin height and orientation. However, this controller requires expensive high-performance hydraulic valves which hinders the applicability to off-the-shelf excavators.
The aim of this project is to leverage RL to deal with the standard, less performant, hydraulic valves. Therefore, the goal is to artificially degrade the performance of the installed valves by introducing delays and dead-zones to converge towards the behavior of standard valves. The controller will be tested on our machine on our test field on the Hönggerberg campus.
References:
[1] https://www.youtube.com/watch?v=5_Eq8CxKkvM
[2] https://www.research-collection.ethz.ch/handle/20.500.11850/120826
To facilitate the operation of walking excavators a chassis balancing controller has been developed that optimizes the wheel contact force distribution given high-level inputs such as cabin height and orientation. However, this controller requires expensive high-performance hydraulic valves which hinders the applicability to off-the-shelf excavators.
The aim of this project is to leverage RL to deal with the standard, less performant, hydraulic valves. Therefore, the goal is to artificially degrade the performance of the installed valves by introducing delays and dead-zones to converge towards the behavior of standard valves. The controller will be tested on our machine on our test field on the Hönggerberg campus.
- Literature review
- Investigate the performance of real pilot-stage valves
- Degrade the actuator performance and train a chassis balancing controller
- Test and evaluate the controller in sim and on the real machine
- Literature review - Investigate the performance of real pilot-stage valves - Degrade the actuator performance and train a chassis balancing controller - Test and evaluate the controller in sim and on the real machine
- High motivation and interest in the topic
- Structured, independent and goal oriented working behavior
- Experience with C++, ROS, PyTorch, RL
- High motivation and interest in the topic - Structured, independent and goal oriented working behavior - Experience with C++, ROS, PyTorch, RL