Julius Hietala
Robotics & Machine Learning
San Francisco, CA
U.S. permanent resident
I’m a robotics and machine learning engineer in San Francisco. My work spans reinforcement learning, sim-to-real transfer, computer vision, and deployment in production.
Experience
Nimble RoboticsSoftware Engineer
Sep 2022–present- Developed and trained reinforcement learning policy architectures, taking models through experimentation and production integration.
- Built the pipeline connecting Python-based reinforcement learning training to production deployment in Rust, keeping model inputs consistent between training and the live system.
- Optimized model inference using GPU execution and TensorRT to reduce deployment latency.
Smartly.ioSoftware Engineer
Jan 2017–Sep 2022- Led development of automated image and video generation and ad optimization solutions.
- Integrated large advertisers with Smartly.io’s machine learning products to scale their advertising workflows.
Stray RobotsCo-Founder, CTO
Apr 2021–Jun 2022- Co-developed RGB-D annotation tools combining SLAM, 3D reconstruction, and label projection to create computer vision training datasets.
- Equal-contribution co-author of “3D Annotation Of Arbitrary Objects In The Wild” (2021 preprint). Contributed to open-source annotation tooling and maintained the Stray Python package.
Aalto UniversityMachine Learning Researcher
Jun 2020–Mar 2022- Trained reinforcement learning policies for dynamic cloth folding and transferred them from simulation to a physical robot using visual feedback and domain randomization.
- Worked across policy learning, C++/ROS control, and MuJoCo simulation to evaluate and improve sim-to-real transfer.
- First-author IROS 2022 paper, a finalist for the Best Paper, ABB Best Student Paper, and Best RoboCup Paper awards.
Education
Aalto UniversityMaster of Science in Technology
2016–2021Major: Machine Learning, Data Science and Artificial Intelligence. Graduated with honors.
Thesis: Dynamic Robotic Cloth Manipulation. Research in the Intelligent Robotics group, supervised by Prof. Ville Kyrki.
Selected publications
A Dataset and Benchmark for Robotic Cloth Unfolding Grasp Selection: The ICRA 2024 Cloth Competition
The International Journal of Robotics Research, 2026. Co-author. Contributed to the competition’s evaluation tooling.
Learning Visual Feedback Control for Dynamic Cloth Folding
IROS, 2022. First author. Best Paper, ABB Best Student Paper, and Best RoboCup Paper finalist.
3D Annotation Of Arbitrary Objects In The Wild
Preprint, 2021. Equal contribution with Kenneth Blomqvist.
Open-source contributions
- Cloth Competition: Evaluation service contributions, including SAM-assisted segmentation, cloth coverage measurement, and dataset/server integration.
- PyTorch ExecuTorch: React Native iOS LLaMA demo and native bridge, merged upstream.
- tch-rs: Tensor-expression fuser configuration binding, merged upstream.
Technical experience
Python · PyTorch · C++ · Rust · TensorRT · MuJoCo · ROS · Core ML · ExecuTorch
Reinforcement learning, visual feedback control, sim-to-real transfer, computer vision, model deployment, and inference optimization.
Additional
Co-founder and board member, Junction (2015–present). Previously President of Aalto Entrepreneurship Society.
IEEE Finland CSS/RAS/SMCS Joint Chapter Best Paper Award 2022, 3rd place (presented in 2024).