kyry.me/blog
Blog
Workshops I ran, talks and press, and technical notes from training humanoid policies.
- Released: SONIC-K1, an open-weights motion tracker for the Booster K1NVIDIA's SONIC ported to the Booster K1 and trained from scratch with under 1% of the compute. Code, weights and the K1 USD are public.
- Vilnius University article about my researchAn article by Vilnius University about my research on humanoid robots at Kent State. Link coming soon.
- Open-sourced: 3D-printable shoes for the Booster K1The two shoes I made for the Humanoid Robot Fashion Challenge, Dutch clogs and a light vine pair, are open source, with print files and a MuJoCo model of the K1.
- Teach a robot your dance: a university-wide workshopA four-hour open workshop at Kent State on humanoid robot learning, end to end, from a phone video to a dance on the real Booster K1. About 13 people came.
- Released: Isaac Lab, end to end on H100An open guide to training Isaac Lab policies headless on H100 GPUs and watching them without any RTX GPU: what works, which errors to ignore, and a ready Unitree G1 USD.
- 1st place: Best Humanoid Wearable/AccessoryOur team won Best Humanoid Wearable/Accessory ($1,000) at the Humanoid Robot Fashion Challenge, Kent State: 45 students from six universities, 36 hours.
- Our K1 at Kent State Preview Day, on Fox 8Represented the ATR Lab at Kent State's Preview Day with our Booster K1. The robot made it into Fox 8's coverage.
- Reinforcement learning workshop for the labA hands-on workshop for the ATR Lab on teaching a humanoid to move with reinforcement learning, from a naive reward to motion tracking, on the Booster K1.
- Humanoid learning workshops for the labTwo workshops on humanoid robot learning for members of the Advanced Telerobotics Research Lab, during the fall 2026 semester.
- Late Breaking Results poster at IROS 2026Presented our poster on learning a catwalk gait for the Booster K1 from a single runway video at IROS 2026 in Pittsburgh.
- Presenting our VLM bias paper at IVUS 2026Presented our study of knowledge bias in open vision-language models at IVUS 2026 in Kaunas.