Hey everyone,
Excited to share a significant update on PHASMA
The project continues its focus on true embodied intelligence: humanoid agents learning directly in a 3D physics-based environment through continuous interaction. No motion capture, no scripted behaviors — only raw reinforcement learning where balance, joint limits, contact forces, and real physics fully shape the emerging policy.
Key Update:
Training and iteration now run locally in the browser. You can jump in immediately, interact with the environment, tweak rewards or conditions on the fly, and watch the humanoid adapt in real time — no login, no server delay, no queue.
Client-side execution delivers fast feedback loops, while you can still optionally sync checkpoints and progress to the Internet Computer for persistence.
This makes experimentation much more accessible and fun.
Live Demo: https://phasma.biz
We’re still early, but the progress has already caught the interest of Neura Robotics, which is very encouraging.
Looking forward to your input!
