The rapid development of deep learning relies on the scale of datasets, models, and computational power. In the fields of natural language processing and computer vision, researchers have already discovered a power-law relationship between model performance and data scale. However, in the field of robotics, especially in robot manipulation, such scalable patterns have yet to be established. A research team from Tsinghua University recently published a paper exploring data scaling laws in robot imitation learning and proposed an efficient data collection strategy that collected sufficient data in just one afternoon, enabling...

The Massachusetts Institute of Technology (MIT) showcased a brand new robot training model this week that abandons the previous focus on training methods centered around specific datasets, opting instead for an approach that utilizes vast amounts of information similar to that used in training large language models (LLMs). Researchers noted that imitation learning—where agents learn by mimicking individuals performing tasks—can fail when facing minor challenges. These challenges may include varying lighting conditions, different environmental setups, or new obstacles. In such cases, the robots lack sufficient data to adapt to these changes.

This week, MIT showcased a new model for training robots, aimed at addressing the challenges faced by imitation learning when introduced to small obstacles. Researchers pointed out that imitation learning can fail in conditions such as varying lighting, different environments, or new obstacles because robots simply do not have enough data to adapt. The team sought powerful data methods, like those used in models such as GPT-4, to tackle the problem. They introduced a new architecture called the Heterogeneous Pretrained Transformer (HPT), which aggregates information from different sensors and environments.

Researchers from Stanford University have proposed a low-cost holistic teleoperation system named Mobile ALOHA for collecting data on holistic teleoperation. Mobile ALOHA extends the capabilities of the original ALOHA by placing it on a wheeled base, giving it mobility. The researchers utilized a static ALOHA dataset for imitation learning, achieving good performance in mobile operation tasks through pre-training and joint training. This study addresses tasks that require holistic teleoperation on a daily basis.

DeepMind has launched a new artificial intelligence agent system that learns tasks directly from human teachers. The system learns through imitation learning in a cultural transmission manner, without relying on human data. The research published in the journal Nature details the process and advantages of imitation learning. This approach has significant implications for training robots and customer service systems. DeepMind addresses data collection and privacy issues in AI training.

Research shows that the MimicGen system can generate large-scale training data for robots. By leveraging digital twin technology, it replicates human operation data in a simulated environment. Using fewer than 200 human demonstrations, it autonomously generates 50,000 training data points. Synthetic data and simulation are crucial for AI development, providing an infinite source of training data. MimicGen is suitable for long-sequence tasks and high-precision tasks, such as grasping and assembling.