Skild AI
Skild AI - Robot Foundation Model Startup from Carnegie Mellon
Skild AI
Home > Companies > Skild AI
Overview
Skild AI is a robotics AI startup founded by Carnegie Mellon University faculty, developing a general-purpose robot “brain.” In 2024, they achieved a $1.5B valuation with a $300M Series A round.
| Item | Details |
|---|
| Headquarters | Pittsburgh, PA |
| Founded | 2023 |
| Co-Founders | Deepak Pathak, Abhinav Gupta |
| Funding | $300M (Series A) |
| Valuation | $1.5B (2024) |
| Investors | Lightspeed, Coatue, SoftBank, Bezos |
Founding Team
Deepak Pathak (CEO)
- Carnegie Mellon Assistant Professor
- Expert in self-supervised learning
- Research on curiosity-driven reinforcement learning
Abhinav Gupta (Chief Scientist)
- Carnegie Mellon Professor
- Expert in robotics and computer vision
- DARPA, NSF research leader
Approach
General-Purpose Robot Brain
┌─────────────────────────────────────┐
│ Skild Foundation Model │
│ "Any Robot, Any Task, Anywhere" │
└─────────────────────────────────────┘
│
┌─────┴─────┐
▼ ▼
┌───────┐ ┌───────┐ ┌───────┐
│Robot A│ │Robot B│ │Robot C│
└───────┘ └───────┘ └───────┘
Core Strategy
| Strategy | Description |
|---|
| Large-scale Simulation | Pre-training in diverse environments |
| Cross-embodiment | Support for various robot forms |
| Foundation Model | Fast adaptation to new tasks |
| General Purpose | No domain limitations |
Simulation-Based Learning
Data Generation
- Building large-scale simulation environments
- Generating diverse physical scenarios
- Applying Domain Randomization
Sim-to-Real Transfer
[Simulation] ──Domain Randomization──→ [Robust Policy] ──Transfer──→ [Real Robot]
Technical Features
Self-Supervised Learning
- Utilizing large-scale unlabeled data
- Learning physics from internet videos
- Curiosity-driven exploration
Scalable Architecture
- Applying LLM scaling laws to robotics
- More data/compute → Better performance
- Cross-robot generalization
Roadmap
| Timeline | Milestone |
|---|
| 2023 | Company founded |
| 2024.07 | $300M Series A |
| 2024-25 | Foundation Model development |
| 2025+ | Partner robot integration |
Differentiators
vs Physical Intelligence
| Factor | Skild AI | Physical Intelligence |
|---|
| Background | CMU academia | Google/Stanford |
| Approach | Simulation-centric | Real data-centric |
| Funding | $300M | $400M+ |
Simulation vs Real Data Debate
Skild: "Infinite data generation possible through simulation"
+ Overcome Sim-to-Real gap with Domain Randomization
Physical Intelligence: "Real physical interaction is essential"
+ Simulation has its limitations
References
See Also