Towards Generic Robot Learning

In recent months, we have seen remarkable advances in large-scale, multi-modal foundation and generative models. These models hold promise for significantly improving robots' ability to acquire diverse skills. Our goal is to deploy these models in a responsible way to enable high-performance robots that can operate effectively in the real world. Specifically, we aim to focus on efficiently implementing generative models on physical robots for real-world data collection and training. This includes leveraging unstructured passive human demonstrations and scaling up structured demonstration collection. Our key advantage is having access to customized, state-of-the-art robots that are optimized for AI training and deployment. By combining advanced generative models with tailored robot platforms, we believe we can achieve major breakthroughs in real-world robot learning and performance.

Key qualifications:

•    Large model algorithm deployment on constricted hardware platforms (language/audio/vision/robot models on a GPU with resources no more than A40 GPU)
•    Hardware acceleration for large models (deep compression, quantization, NAS, knowledge distillation, etc.)
•    Research background in robotics AI is a plus
•    Relevant publications are required for postdoc positions

Contact:
Please send your comprehensive CV to email: eeqshao@ust.hk (only shortlisted candidates will be contacted)