with Jaehyung Kim
If you're looking to contribute to a revolutionary robotics project that addresses the challenges preventing robots from being widely adopted in homes, then look no further! We're currently seeking interns to help us develop a low-sim-to-real gap robot arm for robot learning.
One of the main challenges preventing robots from being widely adopted in homes is the lack of intelligence in robot manipulation that can be employed in the real world. We recognize that achieving sim-to-real transfer is crucial for making intelligent manipulation a reality. To overcome this challenge, we're leveraging recent advances in GPU-based simulators to generate large-scale robot interaction datasets, which will help us to develop impressive applications of robot manipulation. However, current robot arm hardware is not adequately equipped for sim-to-real transfer due to many factors that are difficult to model.
In addition to these challenges, current collaborative arms also have limitations that prevent robots from moving agilely and interacting with humans safely. Heavy hardware makes it difficult to move quickly without damaging the environment or humans. We investigated how to reduce these limitations.
To overcome these limitations, our lab is focused on two main areas:
We will borrow ideas from MIT mini cheetah [1] and other quadruped robots. Similar to the quadruped with linkage-based legs, our goal is to attach heavy motors to the base. We are using quasi-direct drive (QDD) actuators and linkage-based transmission to reduce energy loss.
[2, 3] show similar approaches with timing belt transmission.
In this internship,