School of Engineering Department of Electronic and Computer Engineering 169 Embedded Systems for AI Hardware Supervisor: SHAO Qiming / ECE Student: TEH Wei Jing / CPEG Course: UROP 1000, Summer Previous decades of computing have relied heavily on minimizing the size of transistors, following a trend called Moore’s Law which has plateaued due to physical limitations. Using the existing computing paradigm (deterministic computing), there are problems commonly categorised as NP-hard that requires exponentially growing computational time with increasing complexity. This UROP report gives a glimpse on part of the emerging paradigm (probabilistic computing), in particular the concept of Ising Machine on solving combinatorial optimization problems like number-partitioning and max cut. We will explore its implementation using a particular type of magnetic tunnel junction, leveraging on its stochasticity to form probabilistic bits, the basic building block of this new computing paradigm. Embodied Robotic Arm Systems with AI-Based Control Policies Supervisor: SHAO Qiming / ECE Student: GAN Zesen / ISD Course: UROP 1100, Spring Recent advancements in robotics and machine learning have revolutionized embodied robotic arm systems, enabling sophisticated manipulation tasks through AI-based control policies. This literature review examines state-of-the-art approaches, including diffusion-based neural networks, transformers, Vision-LanguageAction (VLA) models, reinforcement learning, and imitation learning, with a focus on efficiency, performance, generalization, error handling, and user interaction. Key reference projects such as Robotic Diffusion Transformers, OpenVLA, and Mobile ALOHA highlight the integration of these methods in real-world applications. The review identifies research gaps, such as computational efficiency and long-horizon planning, and proposes directions for enhancing robotic systems to achieve robust, user-friendly, and adaptable manipulation capabilities. Embodied Robotic Arm Systems with AI-Based Control Policies Supervisor: SHAO Qiming / ECE Student: GUO Youli / ELEC Course: UROP 1100, Summer The following text will be divided into two main sections, the self-learning section and hands-on practice section. I shall explain in detail how I tackled this challenging project, listing out my steps and thoughts. The former section lays the foundation for the understanding of topics crucial to VLA models and helps me gain proficiency in skills required for analyzing and deploying modern AI models. The latter section depicts how I actually got myself involved in the project. From first trying to deploy DP3 locally and test it in simulations to eventually running the model on a real robotic arm in the lab. As of the time this is written, the model’s performance still leaves much to be desired, however, improvements have been made and potential solutions have been proposed.
RkJQdWJsaXNoZXIy NDk5Njg=