Flexion Brain

Architecture

Multi-layer autonomy stack for humanoids

Key facts

Three-layer autonomy stack (primary Flexion)

Flexion's Nov 20 2025 Series A post by CEO Nikita Rudin describes a reinforcement learning and sim-to-real platform for humanoid robots with a Command Layer (language models for task breakdown and grounding), Motion Layer (vision-language-action model trained mainly on synthetic data), and Control Layer (transformer-based low-latency whole-body control with a modular skill library). The company says it is building the brain, not the body ([Flexion](https://flexion.ai/news/flexion-raises-50m-to-build-the-brain-of-humanoid-robots-at-scale))

Series A terms (primary Flexion)

The same post says Flexion raised $50 million Series A from DST Global Partners, NVentures (NVIDIA's venture arm), redalpine, Prosus Ventures, and Moonfire, following $7.35 million seed from Frst, Moonfire, and redalpine. Use of proceeds listed: expand Zurich R&D, scale compute and robot fleets, establish a U.S. presence, and accelerate commercialization. Treat amounts and investors as company-stated ([Flexion](https://flexion.ai/news/flexion-raises-50m-to-build-the-brain-of-humanoid-robots-at-scale))

Reflect robotics intelligence platform

Flexion describes Reflect v1.0 as a robotics intelligence platform for long-horizon humanoid work, combining mission reasoning, perception, physical execution, and recovery ([Flexion Reflect v1.0](https://flexion.ai/news/flexion-reflect-v1.0)).

Whole-body autonomy stack

Flexion says Reflect combines a vision-language mission controller, a VLA motion layer, reinforcement learning, whole-body control, and FlexComm runtime components ([Flexion Reflect v1.0](https://flexion.ai/news/flexion-reflect-v1.0)).

RGB sim-to-real pipeline

Flexion says its collaboration with Niantic Spatial and NVIDIA uses RGB-camera-derived digital twins, NVIDIA Isaac Sim, and Isaac Lab to train humanoid navigation policies and transfer them from simulation to physical environments ([Flexion, Niantic Spatial, and NVIDIA](https://flexion.ai/news/niantic-spatial-flexion-and-nvidia-closing-the-sim2real-gap-for-humanoids)).

Sim-to-real claim

Flexion reports that its RGB-only policies transferred zero-shot from the simulated digital twin to the physical environment with reliability comparable to conventional depth-based methods; this is a company-stated research result ([Flexion, Niantic Spatial, and NVIDIA](https://flexion.ai/news/niantic-spatial-flexion-and-nvidia-closing-the-sim2real-gap-for-humanoids)).

Developed by (1)

• Flexion Robotics

Common questions

Sources (4)

1. https://flexion.ai/news/flexion-raises-50m-to-build-the-brain-of-humanoid-robots-at-scale ↗
2. https://flexion.ai/news/flexion-reflect-v1.0 ↗
3. https://flexion.ai/ ↗
4. https://flexion.ai/news/niantic-spatial-flexion-and-nvidia-closing-the-sim2real-gap-for-humanoids ↗

Related brains

• Genie Operator-1 (GO-1)5 models
• Aurora Driver4 models
• Tesla FSD-Bot4 models
• Kodiak Driver3 models
• Nuro Driver3 models
• Redwood AI3 models