Yuval Goshen

Co-developed and delivered hands-on laboratory sessions on Task and Motion Planning (TAMP) for the ICAPS 2024 Summer School for doctoral students. The lab was led by Dr. Sarah Keren and TA'ed by Guy Azran and myself.
The week-long summer school preceded the main ICAPS conference in Banff, Canada, where I also participated. We developed interactive Jupyter notebooks for teaching TAMP concepts, which are now publicly available as learning resources for the broader planning community.
This experience was particularly enriching, allowing me to collaborate with and learn from talented students and researchers.
Resources: GitHub Repository | Notebooks

• Apple Excellence Award (2024)
  Awarded for academic excellence and social contribution
  Faculty of Electrical and Computer Engineering
• CRML Outstanding Project Award (2023)
  For my graduation project at the CRML lab.
• Dean's List
  Multiple selections during BSc studies (2019-2023)
  Faculty of Electrical and Computer Engineering

• S3E: Semantic Symbolic State Estimation With Vision-Language Foundation Models (second author) | Paper
• Value of Assistance for Grasping (second author) | Paper
• Work in Progress: Value of Assistive Actions in POMDPs for robotics (my thesis paper)

Teaching Assistant at the Technion for:
Computer Vision (046746) - Spring 2023 to Winter 2024
Robotics and AI (236006) - Winter 2024

Member of the Vision Team for Technion's F1TENTH autonomous racing project. Our team developed a robust opponent tracking system based on the YOLO-World pre-trained model, enabling real-time detection and tracking of competitor vehicles during racing scenarios.
We competed at the official F1TENTH competition at ICRA 2024 , where we demonstrated our autonomous racing capabilities against teams from around the world. Beyond the vision system, I also contributed to developing the autonomous driving algorithms for the vehicle.

Vision Team GitHub

Developed and maintaining the core robotics infrastructure for CLAIR Lab, comprising both physical and simulated environments. The stack integrates vision systems, motion planning, and manipulation controllers, supporting multiple research projects across the lab.
Key components include camera transforms and calibration, object detection and position estimation, manipulation control, and a MuJoCo-based simulation environment that mirrors the physical setup. This infrastructure currently supports several undergraduate and graduate research projects and is being expanded into the lab's general robotics framework.
The system serves as the foundation for my empirical evaluation for my thesis research on Value of Assistance in POMDPs, while enabling diverse robotics research across the lab.
This project won CRML lab's 2023 Outstanding Project Award.

GitHub

In this project we built a simulator for robots to navigate in a maze, and trained them to get from start to goal using motor control.
We used hierarchical agents to separate the long horizon problem to two tasks: walking and navigation.
Project partner: Dor Bitton

GitHub

In this personal mini-project, I collected a set of images of my cat, some of them extracted automatically from videos. I trained a VAE (Variational Auto Encoder) with few additional vision tricks to try to generate new images and to interpolate through the latent space.