November 2, 2024: MIT Advances Robot Training with LLM Approach - MIT researchers unveiled a novel method for teaching robots by emulating large language models, using extensive data to improve adaptability. This approach, termed Heterogeneous Pretrained Transformers (HPT), integrates diverse environmental and sensory information, offering a more robust solution than traditional imitation learning, especially when facing variable conditions. The ultimate goal is to create a universal robot brain applicable without additional training, reminiscent of breakthroughs achieved with large language models. Supported by Toyota Research Institute, this development is part of a broader effort to harmonize robotic learning advancements with platforms like Boston Dynamics hardware.
