@exylos_ai just launched on
@virtuals_io
Most people think robotics issue is hardware, but the real issue is high-quality, structured data
More teams are collecting raw footage, and others are relying on simulation, but neither is enough on its own. The real challenge is turning that data into a structured format that robots can learn from.
@exylos_ai solves the pain-point on structured data.
Not just for successful episodes but also failed attempts and recovery, which are critical for teaching robots how to adapt when things go wrong (e.g. hitting an obstacle and repositioning mid-task).
Exylos turns all of these data (while filtering unusable data) into high-quality and useable, structured Skill Packs that can directly be used for policy training and implementation.
Early Evidence:
- They published 3 sample Skill Pack on Hugging Face (github for robotics), and in roughly a month, already have 28k fully organic downloads.
- Making the datasets top 1% and #6 most trending out of 62k robotics dataset. These are downloads by robotics engineers, utilising them.
HF link:
huggingface.co/ExylosAi
Also collaborating with
@reppo to further solve edge cases via RLHF - further improving quality of their output
Founder -
@vadcrypto has been in the space for 10 years, kept building through market cycles. Understood the problem on robotics data first-hand from his previous experience in VR. A real passion for solving this.
One final point: market is already paying for robotics infrastructure today. Lightwheel closed $100M in orders in Q1 2026 alone. This isn't a "maybe in 5 years" category anymore.
When robotics scales, I think the biggest winners won't just be the hardware companies- they'll be the teams building the data layer that everything else depends on