Rhoda AI Closes $450M Series B, Targets General-Purpose Humanoid at Scale
Rhoda AI, founded in 2023 by former OpenAI robotics researcher Dr. Sarah Chen and ex-DeepMind principal scientist Marcus Webb, has closed a $450 million Series B at a post-money valuation of $3.2 billion. The round was led by Sequoia Capital with participation from Microsoft M12, Samsung Ventures, and Fidelity.
The Diffusion Policy Differentiation
Rhoda's technical approach centers on diffusion-policy-based motion generation — applying the same generative AI architecture behind image diffusion models (like Stable Diffusion) to robot motion planning.
Instead of hand-programmed trajectories or reinforcement learning policies trained on reward functions, Rhoda's system:
- Learns from human demonstrations (5–20 examples per new task)
- Generates smooth motion trajectories as a "denoising" process from random noise to purposeful motion
- Generalizes across novel object configurations without explicit reprogramming
Benchmark performance (from Rhoda's published paper, Nature Machine Intelligence, January 2026):
- Columbia University LIBERO benchmark: 94.3% task success rate (vs 71% for nearest competing method)
- Novel object generalization: 78% success on objects never seen during training (vs 34% for behavior cloning baselines)
- Real-world kitchen manipulation: 89% success rate on 47-task cooking benchmark
The caveat: all benchmarks run at reduced speed (60–70% of maximum robot velocity) to maintain reliability. Industrial deployment at full speed remains in testing.
Hardware Platform
Rhoda's humanoid hardware, the Rhoda-1, specs:
- Height: 1.68 m, Weight: 72 kg
- Payload: 8 kg per arm, 16 kg total
- Hands: 5-finger, 22 DOF, tactile sensor array (5,000 sensing points per hand)
- Onboard compute: NVIDIA Thor SoC (automotive-grade AI chip)
- Battery: 6 hours operational
- Target production price (2027): $85,000–$110,000
The 5-finger hands with dense tactile sensing are Rhoda's hardware differentiator — most competitors use simpler 3–4 finger grippers. For manipulation tasks involving irregular objects (garments, flexible packaging, odd-shaped parts), tactile feedback is critical.
Commercial Deployment Plan
Rhoda has signed letters of intent with 8 manufacturing companies for beta deployments in H2 2026:
- 3 electronics assembly operations (PCB handling, connector insertion)
- 2 pharmaceutical companies (vial handling, lab automation)
- 2 food processing facilities (bakery, meat packing)
- 1 automotive tier-1 supplier (wire harness assembly)
These pilots are structured as no-charge technology demonstrations — Rhoda provides robots at cost in exchange for exclusive data rights during the pilot period. After 6 months, customers can purchase at commercial pricing.
Competitive Positioning
The humanoid field has become crowded. Rhoda's positioning:
| Company | Differentiator | Stage |
|---|---|---|
| Rhoda AI | Diffusion policy, advanced hands | Series B, pre-production |
| Physical Intelligence (π) | Foundation model for manipulation | Series C, software platform |
| Figure AI | Fastest to automotive deployment | Series B, BMW partnership |
| Agility Robotics | Strongest logistics deployment | Commercial, Digit in production |
| Unitree | Lowest price ($16K) | Production, low payload |
Rhoda's bet is that manipulation quality — the ability to handle irregular objects without reprogramming — will be the decisive capability for the widest industrial use cases.
Use of Proceeds
- $180M: Hardware manufacturing scale-up (own facility in Fremont, CA)
- $150M: AI model training compute
- $80M: Customer deployment support and field engineering team (200 hires)
- $40M: International expansion (Japan, Germany)
*Source: Rhoda AI press release, Sequoia Capital announcement, TechCrunch, The Robot Report*