DexFuture: Hierarchical Future-State Visuomotor Targeting for Bimanual Dexterous Tool Use is available on arXiv.
Robotics Researcher & Entrepreneur
Kuang-Ting (Tim) Tu
I work on robotics and computer vision, with a focus on SLAM, robotic manipulation, and pose estimation.
News
Research Internship at VPL and ERL: I am working with Prof. Truong Nguyen's Video Processing Lab (VPL) as a research intern on robotic manipulation and pose estimation research, while also being co-advised by Prof. Nikolay Atanasov in the Existential Robotics Laboratory (ERL).
Show older updates
Master's Program at UCSD: I started my master's studies in UCSD ECE and joined Prof. Nikolay Atanasov's Existential Robotics Laboratory (ERL).
Connextion chapter closed: I wrapped up my work at Connextion and began my next chapter, with deep thanks to every co-founder, teammate, and employee who built alongside me. 🎉🎉🎉
Education
Education
University of California, San Diego, USA
Master of Science in Electrical and Computer Engineering, specializing in Robotics.
Visiting Research Scholar
Existential Robotics Laboratory (ERL), Department of ECE, UC San Diego.
Fully Funded
National Tsing Hua University (NTHU), Taiwan
Bachelor of Engineering in Interdisciplinary Program of Engineering, specializing in Power Mechanical Engineering and Electrical Engineering.
Research
Publications
arXiv Preprint
DexFuture: Hierarchical Future-State Visuomotor Targeting for Bimanual Dexterous Tool Use
- Developed a hierarchical future-state prediction framework that infers sparse hand-tool-object targets from egocentric RGB observations and structured geometric cues, removing the need for privileged future demonstration states during inference.
- Designed a horizon-conditioned transformer with structured visuomotor tokenization and integrated the predicted targets with a physically grounded dexterous control policy for bimanual manipulation.
Portfolio
Selected Projects
Aug 2026
Shared Latent Space Retarget: Embodiment-Agnostic Hand Retargeting for Dexterous Manipulation Transfer
- Built a learned retargeting module that maps human demonstrations and multiple robot hand embodiments into a shared latent pose space using embodiment-specific variational autoencoders.
- Aligned the latent space with a differentiable forward-kinematics contact loss over semantic fingertip pairs, reducing reliance on dense joint-to-joint correspondence or paired human-robot trajectories.
- Evaluated retargeting across Shadow, Inspire, and Allegro hands on single-hand and bimanual dexterous manipulation demonstrations.
June 2023 - Aug 2023
Autonomous Robot Obstacle Avoidance using Depth Sensing
- Simulated in PyBullet and Gazebo to verify whether obstacle avoidance autonomous vehicle can be achieved by combining the conversion of 3D images from a depth camera into a 2D costmap with the AStar and Time Elastic Band (TEB) algorithm.
- Applied simulation results to a real race car and used the Adaptive Monte Carlo Localization (AMCL) method to ensure both systems operated correctly.
Sep 2022 - Apr 2023
Rescue Robot with Autonomous Simultaneous Localization and Mapping (ASLAM), Object Detection and Distributed Computing
- Built an autonomous vehicle to assist in disaster relief using LiDAR and a depth camera to construct a 3D map of the disaster area.
- Reduced the vehicle's GPU load through distributed computing, which allowed minimizing the size of the autonomous vehicle by eliminating the need for high-performance hardware.
- Trained an object detection model to identify human body parts, aiding in the detection of partially buried victims in disaster areas.
Oct 2023 - Dec 2023
Analysis of Basketball Free Throw Techniques Using IMU and Image Recognition
- Constructed a low-cost device for analyzing free throw posture using an inertial measurement unit (IMU), Arduino Nano, SD card module, object recognition model, and 3D-printed components.
- Captured hand trajectory with IMU and recorded the physical parameters of ball, release speed, and angle using ball detection model.
Sep 2021 - Jan 2022
Anti-epidemic-themed Third-person shooter (TPS) Tank Shooting Game
- Developed a virus-shooting game from scratch using C language and the Allegro plugin library. NO AI, NO SHORTCUTS, JUST HAND-CRAFTED CODE AND A LOT OF DEBUGGING.
June 2022
Asia-Pacific Mechanics Contest for College Students (Race car's hub design)
- Designed a racing wheel hub set capable of withstanding maximum pressure while minimizing weight, with the goal of maximizing the down-force/weight ratio.
- Simulated the downward pressure applied to the 3D model on the computer and used the Topology function in Altair Inspire to remove excess unstressed parts to yield a design with optimal down-force/weight ratio.
2022 Asia Pacific Regional Collegiate Student Mechanics Competition Merit Award. Awarded by the sponsoring company CSUN with a pre-employment offer.
Work Experience & Skills
Work Experience
Mar 2026 - Sep 2026
VPL & ERL: Research Intern
Video Processing Lab (VPL) and Existential Robotics Laboratory (ERL), UC San Diego.
- Contributing to Language MISO, a journal manuscript in preparation on multiresolution neural implicit SLAM for scalable scene construction and vision-language spatial understanding.
- Working on hierarchical submap optimization and compact latent feature representations that support both geometric reconstruction and open-vocabulary object grounding.
- Leading an independent DexFuture extension that studies task-conditioned future-state prediction for dexterous manipulation from RGB-D observations, hand-object pose cues, proprioception, and language instructions.
Jan 2025 - July 2025
ITRI: Robotics Engineer
- Industrial Technology Research Institute - Taiwan.
- The main focus is on developing a module that integrates Ultra-wideband (UWB) communication devices with the onboard sensor data of robots, enabling a group of robots to localize each other and form a collaboratively operating robotic network.
- Responsible for developing a novel MCL algorithm by incorporating distance information to estimate the position of each robot.
- In charge of writing libraries for UWB DW3000 module-based ranging and communication.
Nov 2023 - August 2025
Connextion: Founder & CEO
Connextion is my app development company. Its first product, Connection, was built to make campus communication and class coordination smoother.
- Connection facilitates smoother communication between teachers and students within the campus.
- Integrated a chat feature into the course schedule, allowing students to automatically join class discussion groups after setting up their schedule.
- Built and led a team selected for NTHU's Garage incubator.
- Ranked 25th in Taiwan Apple store's social app category in May 2024.
Mapify
Mapify is Connextion's second product, a map-based community for authentic and reliable store reviews.
- Solves the issue of fake reviews on Google Maps by enabling private groups for honest store reviews.
- Helps users build trusted recommendations within their circles without fear of removal.
Skills
Technical Toolkit
Contact
Contact Me
Kuang-Ting Tu(Tim)
kttu@ucsd.edu