About

👋 Hi there! I am Yiyan (Harry) Peng, an M.S. student in Computer Engineering at Northwestern University, where I am advised by Prof. Qi Zhu in the IDEAS Lab. I also collaborate with Professors Ruohan Zhang, Manling Li, Huajie Shao, and Minshuo Chen.

Previously, I received my B.Eng. in Electronic Engineering from the Hong Kong University of Science and Technology (HKUST), where I was advised by Prof. Jun Zhang and Dr. Albert Kai-Sun Wong.

🔬 Research Interests

My research interests lie in robot learning and embodied AI, particularly reliable and adaptive manipulation. I am interested in tasks that require precision or sustained progress over long horizons, and in how policies maintain these capabilities in dynamic manipulation tasks and under inference latency.

My prior work on specification-grounded safety evaluation highlighted the gap between task success and reliable behavior. Building on this experience, I am interested in:

• Representations and learning methods for robot foundation models that generalize across tasks and adapt through interaction, with predictive world models as one possible approach.
• Policy learning and adaptive decision-making for precise and long-horizon manipulation, including how policies use observations, memory, and feedback.

📢 News

• 2026.10:  🎉 GAM accepted to the Pretrain to Adapt workshop at CoRL 2026.
• 2026.09:  🎉 SENTINEL accepted to NeurIPS 2026 E&D as a poster.
• 2026.08:  Released ManiGuard.
• 2025.09:  Started my M.S. in Computer Engineering at Northwestern University.
• 2025.05:  Graduated from HKUST with a B.Eng. in Electronic Engineering.
• 2025.05:  Our Final Year Project received the HKUST ECE Best FYP/T Award 2024-2025, 2nd Runner-Up.

📚 Publications

^* Equal contribution. ^† Project lead.

Grounded Action Model: 3D Grounding as a Foundation for Robotics

Gehao Zhang, Weikai Huang, Shailesh Shailesh, Yiyan Peng, Jiafei Duan, Ranjay Krishna

CoRL 2026 Workshop — Pretrain to Adapt

Website Paper Code

3D-grounded action learning for robotic manipulation, with object-centric representations and a flexible interface for task specification.

ManiGuard: A Benchmark and Data Suite for Specification-Grounded Safety Evaluation and Improvement of Robotic Manipulation

Yiyan Peng^*, Philip Wang^*, Simon Sinong Zhan^*†, Yiqi Lyu, Zhenyang Ni, Jixin Yan, Fiorelli Wong, Ruochen Jiao, Hang Yin, Xinyu Cao, Huajie Shao, Manling Li, Ruohan Zhang, Qi Zhu

arXiv preprint, 2026

Website Paper Code

• A specification-grounded manipulation benchmark with 1,000 evaluation scenarios and 8,000 safety-annotated demonstrations for evaluating and fine-tuning VLA policies.
• Across 23,000+ rollouts, 6–21% of successful simulated executions still violate safety specifications, with evaluation spanning simulation and a real Franka robot.

SENTINEL: A Multi-Level Formal Framework for Safety Evaluation of Foundation Model-based Embodied Agents

Simon Sinong Zhan^*, Philip Wang^*, Justin Liu^*, Yiyan Peng^*, Yiqi Lyu, Zinan Wang, Qineng Wang, Zhian Ruan, Xiangyu Shi, Xinyu Cao, Frank Yang, Zhenyang Ni, Kangrui Wang, Ruohan Zhang, Huajie Shao, Manling Li, Qi Zhu

NeurIPS 2026, Evaluations and Datasets Track (Poster)

Paper Code

• Multi-level safety evaluation for foundation-model embodied agents, using temporal logic to check semantic understanding, action plans, and execution trajectories.
• Localizes safety violations with counterexamples in VirtualHome and AI2-THOR, enabling feedback-driven safety improvement.

💻 Projects

Respiratory Monitoring System

Group Leader, Principal Contributor

HKUST, ECE Final Year Project (FYP), self-proposed

🏆 HKUST ECE Best FYP/T Award 2024-2025 2^nd Runner-Up

Code

Developed MHA-VAE-LSTM for respiratory forecasting and anomaly detection, achieving >95% detection accuracy and ≥10% better reconstruction than VAE-LSTM. Integrated sensors, real-time inference, and ESP-NOW communication into an end-to-end monitoring system with sustained latency below 50 ms.

Vision and Learning for Robotics Manipulation and Grasping

HKUST, ELEC4260, Independent Course Project

Code

Trained and evaluated GraspCNN1 & GraspCNN2 and PoseCNN models for robotic arm manipulation, designing grasp synthesis strategies and validating the grasping pipeline in simulated and real-world settings.

Frontier-based Autonomous Exploration for UGV

HKUST, ELEC4260, Independent Course Project

Code

Implemented LiDAR-based SLAM with ICP alignment, A* path planning, Bezier-curve trajectory generation, and PID control to achieve frontier-based autonomous UGV navigation and exploration.

Autonomous Aerial Robotics

HKUST, ELEC5660, Independent Course Project

Code

Implemented real-time trajectory planning for quadrotors, including tuning PID controller, minimum-snap trajectory generation and obstacle-aware path planning (A*, Dijkstra), to achieve smooth and safe autonomous navigation. Developed vision-based localization and state estimation pipelines: applied PnP pose estimation, stereo visual odometry, and fused IMU & vision data via augmented-state EKF, and validated in simulation and real-world flight tests.

Autonomous and Fast Robots

Cornell University, ECE4160/5160, Independent Course Project

Code

Developed and programmed an autonomous driving framework for a fast robot car from scratch, with sensing (multi-sensor fusion), localization (Bayes filter-based localization), environment mapping, control (PID), path planning, and navigation on a microcontroller board (SparkFun RedBoard Artemis Nano).

🎓 Education

M.S. in Computer Engineering

Northwestern University

Sep 2025 - 2027 (Expected)

B.Eng. in Electronic Engineering

The Hong Kong University of Science and Technology

Sep 2021 - May 2025

Non-Degree Exchange, Electrical and Computer Engineering

Cornell University

Jan 2024 - May 2024

🏆 Honors and Awards

• ECE Best FYP/T Award, 2nd Runner-Up, HKUST
• First Class Honors, HKUST
• Dean’s List, HKUST
• University’s Scholarship Scheme for Continuing Undergraduate Students, HKUST