Shengjia Shao
Papers
2
Total Citations
47
H-Index
2
About
Shengjia Shao is a researcher at the forefront of hardware-accelerated machine learning, with a focused expertise in reinforcement learning (RL) for robotic control. His work bridges the critical gap between theoretical RL algorithms and practical, real-time deployment on embedded systems. Shao’s most impactful contribution, "Towards Hardware Accelerated Reinforcement Learning for Application-Specific Robotic Control" (32 citations), pioneers the use of custom hardware architectures to dramatically speed up RL training and inference, enabling robots to learn and adapt in dynamic environments where traditional software-based approaches fall short. He further advanced the field with "Customised Pearlmutter Propagation: A Hardware Architecture for Trust Region Policy Optimisation" (15 citations), where he designed a specialized accelerator for the computationally intensive Trust Region Policy Optimisation (TRPO) algorithm. This work demonstrated how hardware-software co-design can make state-of-the-art policy optimization feasible for resource-constrained robots. By systematically tackling the latency and power bottlenecks of RL, Shao’s research lays the essential groundwork for next-generation autonomous systems—from agile drones to industrial manipulators—that must make split-second, intelligent decisions without relying on cloud computing.
Research Focus
Key Achievements
Top Papers
- 1
- 2