CV
Education, research experience, industry experience, and skills.
Contact Information
| Name | Can Mizrakli |
| Status | MSc Computer Science Student |
| can.mizrakli@student.kit.edu |
Summary
MSc Computer Science student at KIT focused on imitation learning, state space models, robot manipulation, vision-language-action models, cross-embodiment transfer, and reinforcement learning.
Education
-
2025 - 2027 Karlsruhe, Germany
MSc
Karlsruhe Institute of Technology (KIT)
Computer Science
- Specializations: Human-Centered Machine Intelligence; Robotics and Automation
-
2021 - 2025 Ankara, Turkey
BSc
TED University
Computer Engineering
- Minor in Management
- Secondary focus in Machine Learning
Research Experience
-
Jun 2026 – Present Karlsruhe, Germany
Research Assistant
Computer Vision for Human Computer Interaction (CV:HCI) Lab, Karlsruhe Institute of Technology
- Supporting vision-language-action foundation model adaptation for ARMAR-6/7 humanoids and bimanual Franka Emika platforms through hardware interfacing, teleoperation, and model–robot integration.
- Contributing to research on reducing the manual human supervision required to adapt vision-language-action models to new robot embodiments.
- Employed at the CV:HCI Lab (Prof. Rainer Stiefelhagen); supervised by Dr.-Ing. David Schneider, working within the H²T group.
-
Oct 2025 – Oct 2026 Karlsruhe, Germany
Long-Horizon Imitation Learning with State Space Models
Intuitive Robots Lab, Karlsruhe Institute of Technology
Student Researcher (Seminar & Praktikum) · PyTorch, Mamba
- Investigated whether SSM recurrence, used as a primary temporal backbone rather than a shallow add-on, improves long-horizon, memory-dependent robot manipulation.
- Co-developed ReCAT, a structured recurrent policy combining a hybrid Mamba-attention observation encoder, a six-block recurrent memory, and a flow-matching head that cross-attends to both the current observation and memory in every block.
- Reached 95.3% success on LIBERO, 62.4% on RMBench, and 66.7% on real-robot tasks requiring spatial recall, counting, and timing, versus 8.3% for the strongest baseline with history, with the lowest inference latency among compared memory-based policies.
- Co-first author of the resulting ReCAT paper, supervised by Pankhuri Vanjani and TT-Prof. Dr. Rudolf Lioutikov.
-
2024 – 2025 Ankara, Turkey
Undergraduate Researcher
TED University
TDSal: Task-Based Top-Down Saliency Prediction · Computer Vision, Visual Attention
- Developed a task-driven saliency model integrating YOLO visual features with Sentence-BERT task embeddings through transformer-based multimodal fusion.
- Accepted for publication in the CGI 2026 LNCS proceedings; supervised by Prof. Tolga Kurtuluş Çapın.
Industry Experience
-
Jun 2025 – Sep 2025 Istanbul, Turkey
Artificial Intelligence Engineer
Start You UP
- Built AI instructor models using prompting and fine-tuning workflows for persona-driven interactive learning.
- Contributed to data pipelines for transcription, structured JSON generation, and adaptive question answering using Whisper and Azure OpenAI.
-
Jun 2024 – Aug 2024 Istanbul, Turkey
Software Engineer Intern
Intertech
- Developed a financial literacy chatbot using Azure OpenAI, React.js, and Python.
- Built backend integrations and response-quality evaluation workflows.
Publications
-
2026 ReCAT: Remember, Count, and Time: Structured Recurrent Memory for Robot Manipulation
CoRL 2026 Workshop on Memory for Robot Foundation Models, poster (*equal contribution)
Short workshop version of ReCAT, presented as a poster at CoRL 2026.
-
2026 ReCAT: Remember, Count, and Time: Structured Recurrent Memory for Robot Manipulation
arXiv preprint arXiv:2609.35200 (*equal contribution)
A structured recurrent memory policy for long-horizon robot manipulation requiring spatial recall, counting, and timing.
-
2026 TDSal: A Task-Based Top-Down Saliency Prediction Model
CGI 2026, LNCS Proceedings
A multimodal saliency model conditioned on natural-language task descriptions.
Skills
Programming Languages (Advanced): Python, Java, C, SQL, Julia
Frameworks and Libraries (Advanced): PyTorch, TensorFlow, Hugging Face, OpenCV, YOLO
Tools (Advanced): Git, Linux, Streamlit, LaTeX
Hardware (Advanced): Franka Emika Panda, ARMAR-6/7
Languages
Turkish : Native
English : C1 (IELTS 8.0)
German : A1.2
Spanish : A1