Tzu-Yun Shann

About

Tzu-Yun Shann is a researcher specializing in robot learning, sim-to-real transfer, and imitation learning — areas at the intersection of deep reinforcement learning, computer vision, and autonomous systems. Their most recognized contribution, "Virtual-to-Real: Learning to Control in Visual Semantic Segmentation" (2018), directly tackles one of robotics' most persistent challenges: the reality gap between synthetic training environments and real-world deployment. By leveraging visual semantic segmentation as a bridge between simulated and physical domains, Shann's work offers a practical pathway for training robots safely in simulators before transferring learned behaviors to the real world — a critical advance given the risks and costs of physical data collection. This paper has garnered 69 citations, reflecting its meaningful influence within the robotics and machine learning communities. Shann also explores self-supervised learning through the development of an adversarial exploration strategy for imitation learning, enabling agents to meaningfully explore environments without extrinsic rewards or human demonstrations. Together, these contributions position Shann as a thoughtful contributor to scalable, safer robot learning methodologies, with work particularly relevant to researchers navigating the practical challenges of deploying autonomous systems in unstructured, real-world settings.

Research Focus

Key Achievements

Top Papers

1. 1
   Virtual-to-Real: Learning to Control in Visual Semantic Segmentation

   69 citations · 2018
2. 2
   Virtual-to-Real: Learning to Control in Visual Semantic Segmentation

   7 citations · 2018
3. 3
   Adversarial Exploration Strategy for Self-Supervised Imitation Learning

   2 citations · 2018

Key Collaborators

Contact & Links