Chenyang Cao

University of Toronto PhD student, University of Toronto

I am Chenyang Cao (pronounced as Chan-young Tsao in English), a second year PhD student at the University of Toronto (UofT) advised by Prof. Nicholas Rhinehart. I used to study at Tsinghua University, advised by Prof. Xueqian Wang. I received my B.E. degree in mathematics from Fudan University, and I am lucky to work under the supervision of Prof. Zhenyun Qin and Prof. Hongming Shan. My research interests are reinforcement learning, robotics, continual learning and motion planning. I am also interested in VLA and multimodal perception.

Google Scholar Semantic Scholar GitHub X Resume LinkedIn
Education
  • University of Toronto

    University of Toronto

    PhD in Aerospace Studies & Engineering and Robotics Sep. 2025 - Nov. 2029 (Expected)

  • Tsinghua University

    Tsinghua University

    MS in Electronic and Information Engineering Sep. 2022 - Jun. 2025

  • Fudan University

    Fudan University

    BS in Mathematics and Applied Mathematics Sep. 2018 - Jun. 2022

Honors & Awards
  • First Class Scholarship of Tsinghua University SIGS 2024
  • Second Class Scholarship of Tsinghua University SIGS 2023
  • Meritorious Winner of Mathematical Contest in Modeling 2022
  • Third Class Scholarship of Fudan University 2019-2022
Experience
  • Microsoft

    Microsoft

    Research Intern Oct. 2023 - Mar. 2024

Service
News
2026
🎉 Our paper Push-Wiper is accepted by IROS 2026!
Mar 15
2025
🎉 Our paper AM-FMT is accepted by ROBIO 2025!
Nov 15
🎉 Selected as a reviewer of ICLR 2026!
Oct 09
🎉 Begin my PhD at the University of Toronto!
Sep 01
🎉 My website is successfully built. Welcome to visit!
Jun 27
Selected Publications (view all )
Push-Wiper: Toward General-Purpose Robotic Cleaning across Varied Stains and Surfaces with Segmented Pushing Trajectories
Push-Wiper: Toward General-Purpose Robotic Cleaning across Varied Stains and Surfaces with Segmented Pushing Trajectories

Renhao Lu, Mingxin Wang, Chenyang Cao, Yang Yang, Guoping Pan, Kangkang Dong, Yi Cheng, Houde Liu†(† corresponding author)

IROS 2026 Conference

We reformulate viscous stain cleaning as an aggregation problem, where a sponge progressively gathers the stain through segmented pushing trajectories generated by a diffusion policy, generalizing in a zero-shot manner to unseen stains and curved surfaces.

[Paper] [Project Page]

Push-Wiper: Toward General-Purpose Robotic Cleaning across Varied Stains and Surfaces with Segmented Pushing Trajectories
Push-Wiper: Toward General-Purpose Robotic Cleaning across Varied Stains and Surfaces with Segmented Pushing Trajectories

Renhao Lu, Mingxin Wang, Chenyang Cao, Yang Yang, Guoping Pan, Kangkang Dong, Yi Cheng, Houde Liu†(† corresponding author)

IROS 2026 Conference

We reformulate viscous stain cleaning as an aggregation problem, where a sponge progressively gathers the stain through segmented pushing trajectories generated by a diffusion policy, generalizing in a zero-shot manner to unseen stains and curved surfaces.

[Paper] [Project Page]

QPILOTS：Efficient Test-Time Q-Steering for Flow Policies
QPILOTS：Efficient Test-Time Q-Steering for Flow Policies

Yifan Ruan, Chenyang Cao, Andreas Burger, Ali Pesaranghader, Kaveh Kamali, Jaehong Kim, Nandita Vijaykumar, Alan Aspuru-Guzik, Igor Gilitschenski, Nicholas Rhinehart†(† corresponding author)

arXiv 2026 Conference

We introduce a novel algorithm to combine a critic with a pretrained flow-matching policy to achieve strong results in offline-online reinforcement learning (RL) and online VLA steering in simulation.

[Paper]

QPILOTS：Efficient Test-Time Q-Steering for Flow Policies
QPILOTS：Efficient Test-Time Q-Steering for Flow Policies

Yifan Ruan, Chenyang Cao, Andreas Burger, Ali Pesaranghader, Kaveh Kamali, Jaehong Kim, Nandita Vijaykumar, Alan Aspuru-Guzik, Igor Gilitschenski, Nicholas Rhinehart†(† corresponding author)

arXiv 2026 Conference

We introduce a novel algorithm to combine a critic with a pretrained flow-matching policy to achieve strong results in offline-online reinforcement learning (RL) and online VLA steering in simulation.

[Paper]

AM-FMT: Assisting Metric Fast Marching Tree for Motion Planning in Complex Environments
AM-FMT: Assisting Metric Fast Marching Tree for Motion Planning in Complex Environments

Renhao Lu, Yi Cheng, Chongkun Xia, Chenyang Cao, Lunfei Liang, Xiaojun Zhu, Houde Liu†(† corresponding author)

ROBIO 2025 Conference

We propose an assisting metric for the real-time fast marching tree, which speeds up the tree expansion and avoids the local minima trap in environments with narrow passages and dynamic obstacles.

[Paper]

AM-FMT: Assisting Metric Fast Marching Tree for Motion Planning in Complex Environments
AM-FMT: Assisting Metric Fast Marching Tree for Motion Planning in Complex Environments

Renhao Lu, Yi Cheng, Chongkun Xia, Chenyang Cao, Lunfei Liang, Xiaojun Zhu, Houde Liu†(† corresponding author)

ROBIO 2025 Conference

We propose an assisting metric for the real-time fast marching tree, which speeds up the tree expansion and avoids the local minima trap in environments with narrow passages and dynamic obstacles.

[Paper]

Residual Reward Models for Preference-based Reinforcement Learning
Residual Reward Models for Preference-based Reinforcement Learning

Chenyang Cao, Miguel Rogel-García, Mohamed Nabail, Xueqian Wang, Nicholas Rhinehart†(† corresponding author)

arXiv 2025 Conference

We propose a residual reward model for reward learning by effectively taking advantage of human prior knowledge.

[Paper] [Project Page]

Residual Reward Models for Preference-based Reinforcement Learning
Residual Reward Models for Preference-based Reinforcement Learning

Chenyang Cao, Miguel Rogel-García, Mohamed Nabail, Xueqian Wang, Nicholas Rhinehart†(† corresponding author)

arXiv 2025 Conference

We propose a residual reward model for reward learning by effectively taking advantage of human prior knowledge.

[Paper] [Project Page]

FOSP： Fine-tuning Offline Safe Policy through World Models
FOSP： Fine-tuning Offline Safe Policy through World Models

Chenyang Cao, Yuchen Xin, Silang Wu, Longxiang He, Zichen Yan, Junbo Tan, Xueqian Wang†(† corresponding author)

ICLR 2025 Conference

We propose a safe offline-to-online reinforcement learning algorithm by leveraging world models. It ensures the agent safely moves in the environment during online fine-tuning.

[Paper] [Code] [Project Page]

FOSP： Fine-tuning Offline Safe Policy through World Models
FOSP： Fine-tuning Offline Safe Policy through World Models

Chenyang Cao, Yuchen Xin, Silang Wu, Longxiang He, Zichen Yan, Junbo Tan, Xueqian Wang†(† corresponding author)

ICLR 2025 Conference

We propose a safe offline-to-online reinforcement learning algorithm by leveraging world models. It ensures the agent safely moves in the environment during online fine-tuning.

[Paper] [Code] [Project Page]

A Wristband Haptic Feedback System for Robotic Arm Teleoperation
A Wristband Haptic Feedback System for Robotic Arm Teleoperation

Silang Wu, Huayue Liang, Chenyang Cao, Chongkun Xia, Xueqian Wang, Houde Liu†(† corresponding author)

ROBIO 2024 Conference

We make a wristband teleoperation system to provide haptic feedback from the end-effector.

[Paper]

A Wristband Haptic Feedback System for Robotic Arm Teleoperation
A Wristband Haptic Feedback System for Robotic Arm Teleoperation

Silang Wu, Huayue Liang, Chenyang Cao, Chongkun Xia, Xueqian Wang, Houde Liu†(† corresponding author)

ROBIO 2024 Conference

We make a wristband teleoperation system to provide haptic feedback from the end-effector.

[Paper]

Offline Goal-Conditioned Reinforcement Learning for Safety-Critical Tasks with Recovery Policy
Offline Goal-Conditioned Reinforcement Learning for Safety-Critical Tasks with Recovery Policy

Chenyang Cao, Zichen Yan, Renhao Lu, Junbo Tan†, Xueqian Wang†(† corresponding author)

ICRA 2024 Conference

We propose an offline goal-conditioned reinforcement learning algorithm to solve the planning problem in constrained environments without interacting with them. The algorithm combines the advantages of efficient planning and safe obstacle avoidance, and effectively balances the optimization of both aspects.

[Paper] [Code]

Offline Goal-Conditioned Reinforcement Learning for Safety-Critical Tasks with Recovery Policy
Offline Goal-Conditioned Reinforcement Learning for Safety-Critical Tasks with Recovery Policy

Chenyang Cao, Zichen Yan, Renhao Lu, Junbo Tan†, Xueqian Wang†(† corresponding author)

ICRA 2024 Conference

We propose an offline goal-conditioned reinforcement learning algorithm to solve the planning problem in constrained environments without interacting with them. The algorithm combines the advantages of efficient planning and safe obstacle avoidance, and effectively balances the optimization of both aspects.

[Paper] [Code]

All publications