Kyr (@kyry_D) profile picture

Robotics Researcher @ ATR Lab | Kent State University | Humanoid Robotics & RL

Joined September 2026
Pinned Tweet
The first open-weights motion tracker for small humanoids is here. We brought NVIDIA's SONIC to the Booster K1 and it follows human motions it has never seen. Trained with under 1% of SONIC's compute. Weights and code 👇
3
12
2
79
3,998
Got curious how well our K1 actually knows where it is, so we taped out a hallway and walked it back and forth. Turns out it thinks 12 m is about 10.8 m. Using just its leg joints and IMU we got way closer. Kind of surprised how much the legs alone can tell you.
1
13
498
Teaching a Unitree G1 893 dances. All from AIST++, all trackable zero-shot by NVIDIA SONIC. 10 styles, 2.8 hours. Play any of them in 3D, links below
1
2
43
1,983
Everyone is talking about AI-generated papers at ICLR, but what about arXiv? There are 42 papers released in the past 2 years that have the same name: “TRACE” This year ICLR got 60,000 submissions and everyone is checking them for AI. arXiv got more than 350,000 papers in the last 12 months. I don’t know about the papers but I think AI helped to pick the names 😅
1
299
Someone counted the names of ICLR 2027 papers and found 5 TACOs, 2 LATTEs and 2 MANGOs. So I asked ChatGPT how many arXiv papers are called TRACE. At least 42 since 2024
3
6
739
15. TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning 16. TRACE: A Self-Evolving Skill Bank for Consistent, Limit-Aware LLM Agents 17. TRACE: Training Reasoning Agents for Causal Exploration with Synthesized Rewards 18. TRACE: Memory-Efficient Time Series Class-Incremental Learning via Temporal-Spectral Reconstruction 19. TRACE: Governing Memory Validity in Evolving Multi-Agent Systems 20. TRACE: A Multi-Agent System for Autonomous Physical Reasoning for Seismology 21. TRACE: Trajectory-Aware Comprehensive Evaluation for Deep Research Agents 22. TRACE: Trajectory Reasoning through Adaptive Cross-Step Evidence Aggregation for LLM Agents 23. TRACE: Turn-level Reward Assignment via Credit Estimation for Long-Horizon Agents 24. TRACE: Accountable Agentic Retrieval for Source Discovery in Digital Archives 25. TRACE: Single-Pass Decoding-Trace Risk Localization for Generation Calibration 26. TRACE: Capability-Targeted Agentic Training 27. TRACE for Tracking the Emergence of Semantic Representations in Transformers 28. TRACE: TRansformer-based Attribution using Contrastive Embeddings in LLMs
1
163
29. TRACE: Textual Relevance Augmentation and Contextual Encoding for Multimodal Hate Detection 30. TRACE: Evidence Grounding-Guided Multi-Video Event Understanding and Claim Generation 31. TRACE: Temporal Grounding Video LLM via Causal Event Modeling 32. TRACE: Grounding Time Series in Context for Multimodal Embedding and Retrieval 33. TRACE Back from the Future: A Probabilistic Reasoning Approach to Controllable Language Generation 34. TRACE: Adaptive Trust-Region Bayesian Optimization for Open-Ended Molecular Design 35. TRACE: Adaptive Credibility Reweighting for Robust Bayesian Model Selection with Source Summary Statistics 36. TRACE: Data-Free Text Reconstruction Attacks against Approximate Unlearning in LLMs 37. TRACE: Training-aware Robustness and Challenge Evaluation for Text-Vision Embedding Models 38. TRACE: Trustworthy Reasoning from Acoustic Cues and Evidence 39. TRACE: Tracking Multimodal Clinical Trajectories for Inpatient Risk Prediction 40. TRACE: Governing AI-Assisted Research with Principles from Computer Science 41. TRACE: Transformer Residual Activation Codebook Encoding for Communication-efficient Distributed Edge Inference 42. TRACE: Topology-Aware Reconnection via Prediction-Adaptive Connectivity Enforcement for Thin-Structure Segmentation
124
BREAKING: I’m opening another fabrication shop (machine shop) in the Midwest 🇺🇸 CNC machine milling, water cutting, metal 3D printing and more. Yes - we’ll be making robot parts.
35
8
1
209
9,383
I built an agentic @bot on the go with my early @PlaiPinHQ device. With the Grok Gadget SDK, I say a task, tap Start, and agents build it on my Mac while the gadget tracks every step. Now I can ship an app without touching a keyboard. One voice command → a Pomodoro timer
3
5
3
26
3,653
This is actually pretty cool
I built an agentic @bot on the go with my early @PlaiPinHQ device. With the Grok Gadget SDK, I say a task, tap Start, and agents build it on my Mac while the gadget tracks every step. Now I can ship an app without touching a keyboard. One voice command → a Pomodoro timer
1
1
16
1,038
This is huge! I just saw a demo of an app to vibe design robot clothing. It simulates cloth physics, contact points, thermodynamics, and applies different gaits to different humanoid robots 🤯 Is robot fashion coming?
12
4
2
40
3,795
Every lab tour, our Booster K1 did the same Michael Jackson dance. We got sick of it. So we made K1 Moves: an open dataset of emotes and dances for the K1. Every clip is trackable by SONIC-K1 out of the box, no retraining. Links below 👇
2
1
28
995
58 motions to start: dances, waves, bows and styled walks, all retargeted to K1. We'll keep maintaining and extending it. K1 Moves: huggingface.co/datasets/Kyry… SONIC-K1: kyry.me/sonic-k1/
1
75
Isaac Lab 3.0 lets you choose the physics and rendering backends that best fit your robot learning workload, as well as different visualization approaches to see what your policy looks like while training. For example, the SONIC whole-body controller from NVIDIA runs across different physics engines and transfers to real hardware, demonstrating policy robustness across simulation backends and the real world. (3/7)
3
7
71
4,660
Late night lab vibes
8
221
Built for the real robot. Booster's factory torque limits, a 20 ms delay and soft joints, all at once: it still finishes 95.9% of unseen motions without falling.
1
1
265