Bolívar Enrique Solarte Pardo, Ph.D. ML & Computer Vision Researcher

I'm a machine learning and computer vision researcher at the Industrial Technology Research Institute (ITRI) in Hsinchu, Taiwan, with 6+ years of research and engineering experience in robot perception and navigation. My work spans omnidirectional vision (fisheye and 360-degree cameras), visual SLAM, and learning-based robot reasoning with large language models (LLMs) and vision-language models (VLMs).

I received my Ph.D. from National Tsing Hua University (2024), advised by Prof. Sun Min, in collaboration with Prof. Wei‑Chen Chiu and Dr. Yi‑Hsuan Tsai. Previously, I completed my M.Sc. in Electrical and Electronic Engineering at National Taipei University of Technology (2019) with Prof. Syh‑Shiuh Yeh.

Based in Hsinchu, Taiwan, I’m always open to new ideas and collaborations worldwide. Please feel free to contact me at any time.

Industrial Technology Research Institute
Researcher. 2024 - present
Research Engineer. 2022 - 2024
Research Intern. 2018 - 2019
Mirle Automation
Research Intern. 2021
National Tsing Hua University
Ph.D. Electrical Engineering and Computer Science, 2024
National Taipei University of Technology
M.Sc. Mechanical and Automation Engineering, 2019
University of the Armed Forces
B.Sc. Mechatronics Engineering, 2014

Publications

iFusion: Inverting Diffusion for Pose-Free Reconstruction from Sparse Views
Chin-Hsuan Wu, Yen-Chun Chen, Bolivar Solarte , Yen-Chun Chen, Min Sun
3DV 2025
Project Page [Project Page] ArXiv [ArXiv] Code [Code]
uLayout: Unified Room Layout Estimation for Perspective and Panoramic Images
Jonathan Lee, Bolivar Solarte , Chin-Hsuan Wu, Jin-Cheng Jhang, Yi-Hsuan Tsai, and Min Sun.
WACV 2025
Code [Code] ArXiv [ArXiv]
Self-training Room Layout Estimation via Geometry-aware Ray-casting
Bolivar Solarte , Chin-Hsuan Wu, Jin-Cheng Jhang, Jonathan Lee, Yi-Hsuan Tsai, and Min Sun.
ECCV 2024
Project Page [Project Page] ArXiv [ArXiv] Code [Code] Dataset [Dataset]
360-MLC: Multi-view Layout Consistency for Self-training and Hyper-parameter Tuning
Bolivar Solarte , Chin-Hsuan Wu, Yueh-Cheng Liu, Yi-Hsuan Tsai, Min Sun
NeurIPS 2022
Project Page [Project Page] ArXiv [ArXiv] Code [Code] Dataset [Dataset]
360-DFPE: Leveraging Monocular 360-Layouts for Direct Floor Plan Estimation
Bolivar Solarte , Yueh-Cheng Liu, Chin-Hsuan Wu, Yi-Hsuan Tsai, Min Sun
IEEE Robotics and Automation Letters (RA-L) 2022
Project Page [Project Page] ArXiv [ArXiv] Code [Code] Dataset [Dataset]
Robust 360-8PA: Redesigning The Normalized 8-point Algorithm for 360-FoV Images
Bolivar Solarte , Chin-Hsuan Wu, Kuan-Wei Lu, Yi-Hsuan Tsai, Wei-Chen Chiu, Min Sun.
ICRA 2021
Project Page [Project Page] ArXiv [ArXiv] Code [Code] Dataset [Dataset]
360SD-Net: 360° Stereo Depth Estimation with Learnable Cost Volume
Ning-Hsu Wang, Bolivar Solarte , Yi-Hsuan Tsai, Wei-Chen Chiu, Min Sun
ICRA 2020
Project Page [Project Page] ArXiv [ArXiv] Code [Code]
Cutting Insert and Parameter Optimization for Turning Based on Artificial Neural Networks and a Genetic Algorithm
Bolivar Solarte , Diego Hidalgo, Syh-Shiuh Yeh.
Applied Sciences 2019
Paper [Paper]

Selected Projects

AutoLink 2.0: Ai Moderator for 360-images alignment
GoThru360 - Virtual Tours & Google street views - 2025
This project implemented a 360-camera pose estimation method for aligning camera views using dense feature-matching and epipolar geometry constraints. As a result, the camera pose estimation latency was reduced from approximately 30 seconds to under 1 second, with an accuracy of ~5 degrees. The system was successfully integrated into the new Auto-Link 2.0 – Easy Link tool as an upgraded module for GoThru360’s Ai Auto-Moderator.
Video [Video]
Semantic Map for Robot Reasoning
ITRI - Industrial Technology Research Institute - 2024
This research project aimed to build a semantic map representation that enables robot reasoning. The system processes sequences of hand-held images to identify the most meaningful objects in the scene and then estimates feature embeddings aligned with natural language using LVLMs and LLMs. These feature embeddings, when projected into 3D, allow the robot to follow natural-language instructions while also interpreting geometric relationships and contextual information. The system was implemented in Python and ROS2 and evaluated in both NVIDIA Isaac simulation and real-world scenarios.
iClerk: Using LLM for product dispatching
ITRI - Industrial Technology Research Institute - 2024
In this project, we developed a clerk robot system that utilizes LVLM and LLM models to understand natural language instructions and autonomously dispatch products. The system can recognize implicit information, interpret contextual requirements, and seamlessly add new products without constraints (no predefined labels, bar codes, or QR codes). It was implemented in Python and ROS. The prototype was presented at TAIROS 2024 in Taipei, Taiwan.
Blog [Blog] Video [Video]
Vision Inspection System
ITRI - Industrial Technology Research Institute - 2024
This project focused on a vision inspection system designed to verify the correct positioning of a workpiece before a CMM machine (Coordinate Measuring Machine) began its inspection. It used an RGB camera and relied solely on image processing, without any deep neural networks. The system was implemented in Python and ROS and achieved a precision of ±0.5 mm and a recall above 99%. It was evaluated in a real-world production scenario.
MVL Challenge: Multi-View Layout Estimation
Workshop on Omnidirectional Computer Vision - CVPR 2023
This was the first challenge on self-training multi-view layout estimation, where the goal was to explore solutions that rely solely on multi-view information to estimate scene layouts without human annotations. The challenge was part of the Workshop on Omnidirectional Computer Vision, held in CVPR 2023.
Project Page [Project Page] Dataset [Dataset] Presentation [Presentation]
Autonomous Vision Inspection System
ITRI - Industrial Technology Research Institute - 2023
This project aimed to develop an autonomous inspection system for a manufacturing process with over 100 production stations. The deployed robot used a classical LiDAR-based SLAM algorithm for localization and a inspection system that incorporated an auto-labeled object detector, a Structure-from-Motion map for each station, and a multi-infrared-camera rig. It achieved a precision of ±5 mm and a recall of over 98%. The system was implemented in C++, Python, and ROS.
Stereo VSLAM for an AMR
MIRLE Automation Corporation - 2022
This project aimed to validate a stereo vision–based system as an alternative to a LiDAR-based approach for an AMR (Autonomous Mobile Robot). It was built from scratch using open-source C++ and Python code. The implemented features include scale recovery, self-calibration, visual odometry, loop closure, and custom sensor fusion with an IMU. When tested on a real robot, the system achieved an error margin within 5 cm and a latency of 20 FPS.
Robot Mobile Platform for Patrolling Convenience Stores
ITRI - Industrial Technology Research Institute - 2018
In this project, I was responsible for the perception module of an autonomous mobile platform patrolling convenience stores. My main task was to develop an object detection system trained on custom products. The system ran on the robot’s onboard hardware at around 20 FPS without a GPU. It was implemented using Python, C++, and ROS. This robot was presented in TAIROS 2018, in Taipei, Taiwan.
Servo Control Tuning Using Genetic Algorithms and Gradient Descent
ITRI - Industrial Technology Research Institute - 2017
This project aimed to optimize and fine tune the control parameters of a servo control system under unknown external conditions. Due to the high non-linearity of the parameters space, we used a Genetic Algorithm to find the most promising region space, and then used a Gradient Descent algorithm to fine the final model. The project was developed in C# and Matlab.
Recommendation Letter [Recommendation Letter]