After attending ICRA 2024, I’m sharing my insights on the differences between robotics and AI research. Contrary to common belief, AI advancements, like ChatGPT, don’t directly translate to robotics due to the complexity of sensorimotor tasks. Moravec’s Paradox highlights that skills below conscious awareness, like perception, are harder for AI than reasoning. Hardware limitations in AI learning are debated, with some believing data from physical interaction is necessary. While AI models excel in Internet tasks, robotics demands precision and adaptability, making data quality crucial. Overall, AI methods may enhance robotics, but replicating real-world dynamism remains a challenge. The road to truly versatile robots lies in overcoming data and computation hurdles.
https://harimus.github.io//2024/05/31/motortask.html