Dániel Horváth

I am a research associate, pursuing my Ph.D. at Institute for Computer Science and Control, Budapest Hungary in collaboration with Eötvös Loránd University, Budapest, Hungary, and, as a Campus France scholar, with MINES Paristech in Paris, France. My research focuses on robotics in the fields of reinforcement learning, curriculum learning, transfer learning, and computer vision.

I received the M.Sc. degree with highest honours in mechatronics from the Budapest University of Technology and Economics, Budapest, Hungary, in 2019, spending one semester each at the Technical University of Denmark, Copenhagen, Denmark and at the Otto von Guericke University, Magdeburg, Germany.

Email / CV / Linkedin / Google Scholar / YouTube / ORCiD

HiER: Highlight Experience Replay for Boosting Off-Policy Reinforcement Learning Agents
Dániel Horváth, Jesús B. Martín , Ferenc G. Erdős , Zoltán Istenes , Fabien Moutarde
IEEE Access, 2024
project page / eval video / presentation video / paper / (arxiv)

We propose data exploitation and data collection curriculum learning methods for actor-critic agents to facilitate exploration in sparse reward robotic environments. Our method significantly outperformed the state-of-the-art.

Sim2Real Grasp Pose Estimation for Adaptive Robotic Applications
Dániel Horváth, Kristóf Bocsi, Ferenc G. Erdős , Zoltán Istenes
IFAC-PapersOnLine, 2023; Oral presentation on the 22nd IFAC World Congress
project page / 1-min video / video presentation / paper / (arxiv)

We extended our sim2real object detection transfer learning method to multi-object grasp-pose estimation (MOGPE). MOGPE Real-Time and MOGPE HighPrecision yielded an 80% and a 96.67% success rate in a real-world robotic pick-and-place task.

Object Detection Using Sim2Real Domain Randomization for Robotic Applications
Dániel Horváth, Ferenc G. Erdős , Zoltán Istenes , Tomáš Horváth , Sándor Földi
IEEE Transactions on Robotics, 2023; Oral presentation at ICRA 2023
project page / video presentation / paper / (arxiv)

We propose a sim2real transfer learning method based on domain randomization for object detection to generate labeled synthetic datasets. We could shrink the reality gap and achieved 86.32% and 97.38% mAP₅₀ scores in the case of zero-shot and one-shot transfers.

Campus France doctoral scholarship. I was awarded the Campus France scholarship by the French Government to pursue my doctoral research topic, under the supervision of Fabien Moutarde at the École nationale supérieure des mines de Paris (Mines Paris-PSL) , France. I have had close collaboration with Jesús B. Martín working on reinforcement learning in robotics.

2022-2023

Campus Mundi master exchange scholarship. I was awarded the Campus Mundi Scholarship to spend a semester at the Technical University of Denmark in Copenhagen. During my tenure, my studies focused on machine learning and computer vision.

2018

Erasmus+ master exchange scholarship. I was awarded the Erasmus+ to spend a semester at the Otto von Guericke University in Magdeburg, Germany. During my tenure, my studies focused on reinforcement learnin and database concepts.

2018

New National Excellence Program // Új Nemzeti Kiválósági Program (ÚNKP). I was awarded the research grant from the New National Excellence Program to support my work on multi-modal perception for collaborative robot arms. Issued by the National Research Development and Innovation Office of Hungary.

2017-2018

Hungarian National Higher Education Scholarship // Nemzeti Felsőoktatási Ösztöndíj. The new version of the Hungarian Republican Scholarship. The Minister for Education grants the Hungarian National Higher Education Scholarship based on a ranking suggested by the higher education institutions. This scholarship is awarded to 0.8% of the institution's full-time students. Issued by the National Research Development and Innovation Office of Hungary.

2017-2018

Hungarian Republican Scholarship // Köztársasági Ösztöndíj. The former version of the Hungarian National Higher Education Scholarship. The Minister for Education grants the Hungarian Republican Scholarship based on a ranking suggested by the higher education institutions. This scholarship is awarded to 0.8% of the institution's full-time students. Issued by the National Research Development and Innovation Office of Hungary.

2016-2017

BME Faculty of Mechanical Engineering Scholarship // BME Gépészmérnöki Kari Ösztöndíj. Awarded by the Mechanical Engineering Faculty (GPK) of the Budapest University of Technology and Economics in Hungary, the scholarship is intended for students who are actively involved in academic and scientific endeavors that exceed the expectations of their standard curriculum.

2016-2017

I was honored with the HUN-REN SZTAKI Publication Award for my papers on sim2real transfer learning in robotics, published in 2023. I am deeply grateful to my co-authors, Ferenc G. Erdős , Zoltán Istenes , Tomáš Horváth , Sándor Földi , and Kristóf Bocsi, as well as to the colleagues and friends who helped me in my research.
📰 Object Detection Using Sim2Real Domain Randomization for Robotic Applications
📰 Sim2Real Grasp Pose Estimation for Adaptive Robotic Applications

2024

Young Author Award at the INCOM 2021 Symposium, Finalist

I presented our work ( Ferenc G. Erdős , Dániel Horváth, Gergely Horváth) titled 'Visual servo guided cyber-physical robotic assembly cell' at the 17th IFAC Symposium on Information Control Problems in Manufacturing. In this work, we propose a cyber-physical-system-based framework of a robotic assembly cell that is capable of recognizing resources of assembly processes, based on visual input and updating and adapting the assembly plan in the digital twin of the robotic cell.

2021

BME GPK Scientific Students' Conference, 3rd place
Dániel Horváth

I presented my work on multi-modal sensing for adaprive robotic manipulation tasks at a smart factory environment of SZTAKI. The robot needed to perform a cube-building task relying on visual and force feedback. Furthermore, the oparator could specify the the target in a simulation environment and supervise the process with its gestures detected by a Kinect camera. The research was in part funded by the New National Excellence Program.

2017

XXXIII. Hungarian National Scientific Students' Conference, 1st place
BME GPK Scientific Students' Conference, 1st place
Audi Hungaria Special Award
Dániel Horváth, Tamás Magyar, Gábor Losonczi

We presented our work on the development and implementation of navigation and control algorithms for automated guided vehicles at the smart factory environment of SZTAKI. Our tasks ranged from low-level motor control to high-level path planning, scheduling, and task sharing in a multi-robot environment.

2016-2017

World Robot Olympiad, Hungarian National Level , 4th place
Dániel Horváth, Bálint Borosnyay, Dániel Uzseka

As bachelor students, we participated at the Hungarian National Level of the World Robot Olympiad. The aim of the competition was to design and program a robot capable of participating in a unique bowling game variant. Although the balls and pins are smaller than standard, the challenge is heightened by the requirement to retrieve the balls from a specific container and the pins being placed in random patterns, occasionally with obstacles present. Our responsibilities ranged from mechanical and electrical engineering to software development, encompassing a computer vision algorithm for detecting the pins.

2016

BME MOGI Micromouse Robotic Cometition, Software Category , 1st place
Dániel Horváth, Bálint Borosnyay, Dániel Uzseka

As bachelor students, we participated at the software section of the Micromouse Robotic Competition, held by the Mechatronics Department of the Budapest University of Technology and Economics. The objective of the competition is twofold. First, the target position within the maze is needed to be found. Subsequently, the robots were given several attempts to navigate from the starting point to the target position in the shortest time possible. Our tasks ranged from low-level PID control to high-level path planning and graph search.

2016

Reinforcement Learning in Robotics

• Tabular RL: dynamic programming, monte carlo, and temproal difference methods such as SARSA, Q-learning, and Expected SARSA, following (Sutton and Barto 2018).
• Continuous RL: DQN, Cross-Entropy, REINFORCE, A2C, A3C, PPO, TRPO, ACKTR DDPG, SAC, and TD3.
• Curriculum learning methods: HER, PER, HiER.

Last update: 10th of Apr, 2024

I recently had the opportunity to attend the ERCIM "Beyond Compliance 2024: Research Ethics in the Digital Age". As a computer scientist, I found it extremely valuable to learn directly from experts about the ethical challenges of AI. 🙏 I want to express my gratitude to all the participants, as well as to András Benczúr for providing this opportunity and organizing such a valuable event.

Links: Linkedin 🇬🇧

Oct, 2024

Yesterday, I was honored with the HUN-REN SZTAKI Publication Award⚡for my papers on sim2real transfer learning in robotics 🤖, published in 2023 📚. I am deeply grateful to my co-authors, Ferenc G. Erdős , Zoltán Istenes , Tomáš Horváth , Sándor Földi , and Kristóf Bocsi, as well as to the colleagues and friends who helped me in my research.
📰 Object Detection Using Sim2Real Domain Randomization for Robotic Applications
📰 Sim2Real Grasp Pose Estimation for Adaptive Robotic Applications

Links: Linkedin 🇬🇧

Jun, 2024

Last week, I had the privilege of presenting at the 2024 IVSZ - IT Association of Hungary SMART Conference. My presentation explored the application of reinforcement learning (#RL) in robotics 🤖, alongside discussions on large language models (#LLMs) and joint embedding predictive architectures (#JEPA). I am grateful to Peter Smejkál, Kaiser Edit, and Balazs Horvath for the invitation and for organizing such a fantastic event.

Links: SZTAKI 🇭🇺, Linkedin 🇬🇧

Apr, 2024

We had a fantastic time at the 2nd Academic Exchange visit to the Technische Universität Wien. During the two-day workshops, we had fruitful discussions of the current trends in robotics and manufacturing science. Our group focused on human-centered robotics and machine learning. Many thanks to Fazel Ansarin and Sebastian Schlund for organizing this incredible event.

Links: Linkedin 🇬🇧

Apr, 2024

We had the privilege of exploring the cutting-edge projects and research at the Laboratory for Robotics and Intelligent Control Systems (LARICS). We also had the opportunity to showcase our own research and, alongside Nelli Nyisztor and Zoltán Istenes, conducted an engaging workshop on reinforcement learning in robotics 🤖. Our time in Zagreb 🇭🇷 was truly inspiring, and we extend our heartfelt gratitude to Tamara Petrović, Stjepan Bogdan, and Ana Milas for their exceptional programs and warm hospitality.

Links: Linkedin 🇬🇧

Jan, 2024

I presented 👨‍🏫 my work on reinforcement learning and sim2real transfer learning in robotics 🤖 at the Workshop Day of the MILAB (the Artificial Intelligence National Laboratory). These events help to meet with the Hungarian 🇭🇺 machine-learning community, share knowledge, make connections, and start collaborations.⭐

Links: MILAB 🇭🇺, Linkedin 🇬🇧

June, 2023

I participated in the IEEE ICRA23 conference held in London 🇬🇧. As a young scientist👨‍🔬, it was an amazing opportunity to follow the interesting talks at the conference, watch the poster presentation sessions, get to know the newest inventions of the research labs, and share our knowledge in the robotics community 🤖

I also presented my IEEE Transactions on Robotics paper titled Object Detection Using Sim2Real Domain Randomization for Robotic Applications.

Links: SZTAKI 🇭🇺, Linkedin 🇬🇧

June, 2023

I participated in the 2023 Future: Hungary 🇭🇺 conference in London 🇬🇧 as a panelist with Anna Takacs and Emese Thamó in the AI and emerging technologies section. 🎤🤖 We had an engaging discussion on the latest advancements in AI and their impact on business and society. I am grateful for the opportunity to exchange our knowledge, attend fascinating talks, and meet many ambitious Hungarians of different domains.

Links: SZTAKI 🇭🇺, Linkedin 🇬🇧

Apr, 2023

The new volume of the SZTAKI EMI Magazine 🇭🇺, is out. Among other interesting topics, you can find a short interview with me. Thank you Peter Smejkál for the opportunity! 📰🤖

Links: SZTAKI EMI Magazine 🇭🇺, Linkedin 🇭🇺 / 🇬🇧

Oct, 2022

I am joining CAOR: Centre for Robotics Laboratory 🤖 at the MINES ParisTech, Université PSL 🇫🇷 for nine months, focusing primarily on reinforcement learning and sim2real transfer in robotic manipulation tasks under the supervision of Prof. Fabien Moutarde. 📚

Links: Linkedin 🇬🇧

July, 2022

As a unique celebration of the 20-year-anniversary collaboration between SZTAKI and the Fraunhofer-Gesellschaft, we had a 3-day long open-lab exhibition in Stuttgart 🇩🇪 at the II. Science Festival where we showed our live robotic applications to the audience and also presented workshops in different domains of artificial intelligence. Thank you everyone for the hard work!

Links: SZTAKI 🇭🇺, SZTAKI video 🎥, Linkedin 🇬🇧

June, 2022

We created an robotic-art exhibition called Robotschool and Friends. This work is the result of an intensive collaboration of 4+ weeks between the German artist Ottonie Roeder (Goethe Institute) and our lab, the Engineering and Management Intelligence Laboratory at SZTAKI. I participated as the technical leader of the project, and I have gained valuable experience. This was the first art project I was involved in, and I found the whole process fascinating. I would like to thank everyone who helped to make this project successful.

Links: Népszava 🇭🇺, SZTAKI 🇭🇺

Sept, 2021

I participated in the 2021 East European Machine Learning Summer School focusing on deep learing, transfer learning, computer vision, and robotics. There were many interesting lectures, projects, and online social activities. I gained valuable knowledge and met many fellow researchers.

Links: Presentation 🇬🇧, Linkedin 🇬🇧

July, 2021