Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
DigGPT: Large Language Models for Excavation Planning
Large language models (LLMs) have shown the first sparks of artificial general intelligence. We want to test if GPT 4.0 can solve excavation planning problems.
Keywords: GPT, Large Language Models, Robotics, Deep Learning, Reinforcement Learning
Large language models have shown the first sparks of general artificial intelligence. We want to use a large language model like GPT 4.0 or a distilled version like Alpaca 30B to solve excavation planning problems. This kind of problem can be abstracted into a grid-world and easily rendered in text. The agent receives a target map and has to move around blocks of soil to achieve the goal. See the animation for an agent trained with RL, playing the game. We should test if GPT 4.0 can solve the problem zero-shot or if it needs finetuning in form of supervision or reinforcement learning.
Possible final deployment in the real world with our excavator robot.
Large language models have shown the first sparks of general artificial intelligence. We want to use a large language model like GPT 4.0 or a distilled version like Alpaca 30B to solve excavation planning problems. This kind of problem can be abstracted into a grid-world and easily rendered in text. The agent receives a target map and has to move around blocks of soil to achieve the goal. See the animation for an agent trained with RL, playing the game. We should test if GPT 4.0 can solve the problem zero-shot or if it needs finetuning in form of supervision or reinforcement learning. Possible final deployment in the real world with our excavator robot.
- pipeline to make the LLM play the game by figuring out the right prompts and the rendering of the game
- finetuning with supervised learning or reinforcement learning to teach the model how to play
- real-world deployment to solve excavation tasks
- pipeline to make the LLM play the game by figuring out the right prompts and the rendering of the game - finetuning with supervised learning or reinforcement learning to teach the model how to play - real-world deployment to solve excavation tasks
- general programming experience with python
- experience training neural network
- bonus: experience with large language models
- general programming experience with python - experience training neural network - bonus: experience with large language models