Mechanical search (MS) in cluttered environments remains a significant challenge for autonomous manipulators, requiring long-horizon planning and robust state estimation under occlusions and partial observability. In this work, we introduce XPG-RL, a reinforcement learning framework that enables agents to efficiently perform MS tasks through explainable, priority-guided decision-making based on raw sensory inputs. XPG-RL integrates a task-driven action prioritization mechanism with a learned context-aware switching strategy that dynamically selects from a discrete set of action primitives such as target grasping, occlusion removal, and viewpoint adjustment. Within this strategy, a policy is optimized to output adaptive threshold values that govern the discrete selection among action primitives. The perception module fuses RGB-D inputs with semantic and geometric features to produce a structured scene representation for downstream decision-making. Extensive experiments in both simulation and real-world settings demonstrate that XPG-RL consistently outperforms baseline methods in task success rates and motion efficiency, achieving up to 4.5x higher efficiency in long-horizon tasks. These results underscore the benefits of integrating domain knowledge with learnable decision-making policies for robust and efficient robotic manipulation.
RSS 2025 Workshop on Learned Robot Representations (RoboReps) (NEW)
Underwater simulators offer support for building robust underwater perception solutions. Significant work has recently been done to develop new simulators and to advance the performance of existing underwater simulators. Still, there remains room for improvement on physics-based underwater sensor modeling and rendering efficiency. In this paper, we propose OceanSim, a high-fidelity GPU-accelerated underwater simulator to address this research gap. We propose advanced physics-based rendering techniques to reduce the sim-to-real gap for underwater image simulation. We develop OceanSim to fully leverage the accelerated computing advantages of GPU and achieve real-time imaging sonar rendering and fast synthetic data generation. We evaluate the capabilities and realism of OceanSim using real-world data to provide qualitative and quantitative results.
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025
featured by NVIDIA Robotics
This work addresses the task of modeling Deformable Linear Objects (DLOs), such as ropes and cables, during dynamic motion over long time horizons. This task presents significant challenges due to the complex dynamics of DLOs. To address these challenges, we propose differentiable Discrete Elastic Rods For deformable linear Objects with Real-time Modeling (DEFORM), a novel framework that combines a differentiable physics-based model with a learning framework to model DLOs accurately and in real-time. The performance of DEFORM is evaluated in an experimental setup involving two industrial robots and a variety of sensors. A comprehensive series of experiments demonstrate the efficacy of DEFORM in terms of accuracy, computational speed, and generalizability when compared to state-of-the-art alternatives. To further demonstrate the utility of DEFORM, we integrate it into a perception pipeline and illustrates its superior performance when compared to the state-of-the-art methods while tracking a DLO even in the presence of occlusions. Finally, we illustrate the superior performance of DEFORM when compared to state-of-the-art methods when it is applied to perform autonomous planning and control of DLOs.
Conference on Robot Learning (CoRL) 2024
Manipulating deformable linear objects (DLOs) is challenging due to their complex dynamics and the need for safe interaction in contact-rich environments. Most existing models focus on shape prediction alone and fail to account for contact and tension constraints, which can lead to damage to both the DLO and the robot. In this work, we propose a certifiably safe motion planning and control framework for DLO manipulation. At the core of our method is a predictive model that jointly estimates the DLO's future shape and tension. These predictions are integrated into a real-time trajectory optimizer based on polynomial zonotopes, allowing us to enforce safety constraints throughout the execution. We evaluate our framework on a simulated wire harness assembly task using a 7-DOF robotic arm. Compared to state-of-the-art methods, our approach achieves a higher task success rate while avoiding all safety violations. The results demonstrate that our method enables robust and safe DLO manipulation in contact-rich environments.
ICRA 2025 Workshop on Learning Meets Model-Based Methods for Contact-Rich Manipulation
The goal of this project is to create a real-time path planner for bipedal robots Digit and Cassie, allowing them to travel autonomously and safely without collision with any obstacles. My main contributions are:
Utilized the idea of “template and anchor” to simplify the dynamics of bipedal robots Cassie and Digit by approximating the full-order model (anchor) with a reduced-order model (template).
Solved a sum-of-squares (SOS) optimization problem to outer approximate the forward reachable set (FRS) of the anchor model using the dynamics of the template model with the error between two models being conservatively considered.
Achieved real-time path planning by solving a nonlinear optimization problem inspired by Model Predictive Control (MPC).
Achieved collision-free navigation for Cassie in a simulated warehouse environment, and for Digit in our lab and outside around the Michigan Robotics building.
Since Digit’s low-level controller is proprietary, I am currently developing our own controller to enable Digit to navigate itself in more challenging scenarios
Autonomous Surface Vehicle (ASVs) have extensive application potentials in unmanned hydrographic surveys and marine operations. However, traditional path-following controllers of ASVs may not detect obstacles on the water's surface, such as floating buoys and reefs, which can create serious safety risks. I created a fast, reliable sensing system to address this issue. To accurately locate obstacles on the water's surface, I implemented a sensor fusion method. Rather than the traditional template matching method, I created a novel bag-of-visual-words model to classify obstacles, which significantly improved the speed of the perception algorithm. Eventually, I created an algorithm for safe navigation and applied it to detect an underwater target.
The goal of this project was to do localization for the mobile robot PR2 on the OpenRAVE using different filters, evaluate and compare their performance in terms of the robustness in the environment with obstacles and noise. I created algorithms for filters, robot dynamic functions, sensors functions and path planning. To evaluate the performance of my algorithms, I conducted experiments on several interesting scenarios and different maps, including an open area environment, an environment with obstacles, and an office-like environment. I found that the Extended Kalman Filter (EKF) was more efficient than the Particle filter while the Particle filter performed more robust to the highly nonlinear dynamics and complex environments with many obstacles.
In this project, I created machine learning classifiers for predicting patient outcomes in the peritoneal dialysis (PD) treatment. The dataset that I used in this work was originated from the BRAZPD database. While handling the dataset, I created algorithms for feature selection, data balancing, feature clustering and averaging. I tested the performance of XGBoost model and neural network model that were used for classification. The accuracy of classifiers was quantified by F1 score. From results, important features that have obvious influence on the outcomes were identified.
The goal of this project was to conduct depth prediction for monocular camera image. We achieved this task via building two deep neural networks based on the DenseNet and MiniNet. NYUv2 dataset was utilized for our training and evaluation. The performance of the networks was evaluated in both qualitative and quantitative ways.