Register now After registration you will be able to apply for this opportunity online.
Autonomous Multi-Task Excavation with Transformers
We want to develop a generalist digging agent that is able to do multiple tasks, such as digging and moving loose soil, and/or control multiple excavators. We plan to use decision transformers, trained on offline data, to accomplish these tasks.
Our approach involves using offline reinforcement learning to train transformers, which has been shown to produce generalist agents that can handle multiple tasks [1]. In this thesis, we will focus on using decision transformers [2, 3] to teach our excavator various tasks. The offline data we use can come from a variety of sources, including human demonstrations, previously trained reinforcement learning agents, and traditional control policies. We want to develop a generalist digging agent that is capable of performing multiple tasks, such as digging, dumping and moving loose soil, using our excavator.
The project will be jointly supervised by the RSL and Gravis Robotics. Gravis Robotics is an ETH spin-off from the RSL working on the automation of heavy machinery.
[1] A Generalist Agent, Deepmind
[2] Decision Transformer: Reinforcement
Learning via Sequence Modeling
[3] Multi-Game Decision Transformers
Our approach involves using offline reinforcement learning to train transformers, which has been shown to produce generalist agents that can handle multiple tasks [1]. In this thesis, we will focus on using decision transformers [2, 3] to teach our excavator various tasks. The offline data we use can come from a variety of sources, including human demonstrations, previously trained reinforcement learning agents, and traditional control policies. We want to develop a generalist digging agent that is capable of performing multiple tasks, such as digging, dumping and moving loose soil, using our excavator. The project will be jointly supervised by the RSL and Gravis Robotics. Gravis Robotics is an ETH spin-off from the RSL working on the automation of heavy machinery.
[1] A Generalist Agent, Deepmind [2] Decision Transformer: Reinforcement Learning via Sequence Modeling [3] Multi-Game Decision Transformers
- Train transformers using offline reinforcement learning
- Train skill policies with RL in simulation
- Design of experimental campaign and data collection on real the excavator
- Deployment
- Train transformers using offline reinforcement learning - Train skill policies with RL in simulation - Design of experimental campaign and data collection on real the excavator - Deployment
- Experience in training neural networks and RL
- Experience with ROS and good knowledge of C++
- Experience in training neural networks and RL - Experience with ROS and good knowledge of C++
