Hanwen Ren

I am a Ph.D. student in the Department of Computer Science at Purdue University. Currently, I am working as a Research Assistant at Purdue Cognitive Robot Autonomy & Learning (CoRAL) Lab directed by Prof. Ahmed H. Qureshi.

Before coming to Purdue, I got my B.S. from the UM-SJTU Joint Institute at Shanghai Jiao Tong University and Sc.M. from the Electrical Sciences & Computer Engineering Department at Brown University.

Email: ren221 AT purdue DOT edu

Research

My research explores how robots can intelligently interact with complex environments, focusing on active visual perception, object rearrangement planning, and object manipulation in cluttered environments. I develop learning- and planning-based methods that allow robots to reason about their surroundings and perform multi-step manipulation tasks efficiently.

Multi-Agent Monte Carlo Tree Search for Makespan-Efficient Object Rearrangement in Cluttered Spaces

We introduce CAM-MCTS (Centralized Asynchronous Multi-Agent Monte Carlo Tree Search), a planning framework for collaborative object rearrangement in cluttered environments. CAM-MCTS enables multiple robots to coordinate manipulation actions while optimizing the overall task makespan. The method combines centralized planning with asynchronous execution so agents can begin new actions without waiting for global synchronization, reducing idle time and improving efficiency. The performance are verified in both simulation and real-robot systems.

Integrating Active Sensing and Rearrangement Planning for Efficient Object Retrieval from Unknwon, Confined, Cluttered Environments

We introduce an integrated framework that smartly combines multi-stage active sensing and MCTS-based retrieval planning for object retrieval in unknown, cluttered, confined environments. Our OR-MCTS planner relocates task-blocking objects while guiding sensing toward critical unseen regions. Experiments in simulation and real robots show improved success rate and efficiency over SOTA methods.

Multi-Stage Monte Carlo Tree Search for Non-Monotone Object Rearrangement Planning in Narrow Confined Environments

In this work, we propose a Multi-Stage Monte Carlo Tree Search (MS-MCTS) method leveraging an intelligent subgoal-focused tree expansion algorithm to find high-quality plans for complex non-monotone object rearrangement planning problems in confined environments. Our approach results in near-optimal solutions for various object rearrangement instances of diverse difficulty levels.

Multi-Stage Monte Carlo Tree Search for Non-Monotone Object Rearrangement Planning in Narrow Confined Environments

In this work, we propose a Multi-Stage Monte Carlo Tree Search (MS-MCTS) method leveraging an intelligent subgoal-focused tree expansion algorithm to find high-quality plans for complex non-monotone object rearrangement planning problems in confined environments. Our approach results in near-optimal solutions for various object rearrangement instances of diverse difficulty levels.

Multi-Stage Monte Carlo Tree Search for Non-Monotone Object Rearrangement Planning in Narrow Confined Environments

In this work, we propose a Multi-Stage Monte Carlo Tree Search (MS-MCTS) method leveraging an intelligent subgoal-focused tree expansion algorithm to find high-quality plans for complex non-monotone object rearrangement planning problems in confined environments. Our approach results in near-optimal solutions for various object rearrangement instances of diverse difficulty levels.

Neural Rearrangement Planning for Object Retrieval from Confined Spaces Perceivable by Robot's In-hand RGB-D Sensor

This paper presents a neural network-based object retrieval framework that efficiently performs rearrangement planning of unknown, arbitrary objects in confined spaces to retrieve the desired one. Our method demonstrates high performance by ensuring the relocation of non-target objects clear the way for the robot path homotopy to the given target object, thus significantly increasing the underlying motion planner's efficiency.

Robot Active Neural Sensing and Planning in Unknwon Cluttered Environments

In this work, we present the active neural sensing approach that generates the kinematically feasible viewpoint sequences for the robot manipulator with an in-hand camera to gather the number number of observations needed to reconstruct the underlying unknown cluttered environments. Our results exhibit high performance compared to traditional baselines regarding the number of viewpoints, scene coverage success rates, and planning time.

Cograsp: 6-DOF Grasp Generation for Human-Robot Collaboration

In this paper, we propose a novel, deep neural network-based method called CoGrasp that enables robots to grasp various objects in a human-aware manner by contextualizing human preference. Our user study indicates that our approach allows safe, natural, and social-aware human-robot co-grasping experience.