Neuroplastic AI: the transformer moment for the physical world.
Neuroplastic Neural Networks: incredibly small, elegant and fast.
We've built something different. Our Adaptive Neural Controllers learn the physics of a system from first principles and adapt autonomously in real time. No retuning, no cloud, no compromise on speed. They run on the constrained hardware that's already out there in millions of devices, at timescales that make conventional Deep Learning approaches look Big & Slow.
Traditional Deep Learning: Big & Slow
Dense traditional deep learning techniques conform to rigid networks, unable to adapt and requiring huge computation.
Built using rigid architectures
Data wrangling & filtration
Requires massive sets of operational data
Neuroplastic Neural Network: Small & Fast
Sparse, lightweight neural networks perfectly formed to fit in even the most constrained environments.
Grown not built, using algorithms that mimic nature
Neuroplastic synapses that allow adaptation.
Trained on physics models.
The 4 steps to create an Adaptive Neural Controller
Twinned from physics
Domain Randomization applied
Test in Simulation
Ready to deploy in any hardware
The Highest Intelligence-per-byte AI
We’ve benchmarked Luffy AI’s ANCs against the Google DeepMind Real World RL Suite.
The result is 800 x fewer synapses, 400 x less compute required for the equivalent or better performance in tasks.
This cartpole test simulates a real-world physical example using Luffy AI’s ANC. Read more here.
Comparable Performance
Capability per kFLOP
“An empirical investigation of the challenges of real-world reinforcement learning”, G. Dulac-Arnold et al., 2020, arXiv:2003.11881
Fits Anywhere.
Nano AI
Microcontrollers, Embedded Boards, Variable Frequency Drives, Firmware etc
>100Hz
<1kB.
Next-gen intelligent devices
Micro AI
PLC, DCS, SCADA, EDGE
10-100Hz
1- 25kB.
Application-based optimisation
Mini AI
SCADA, Supervisory, Modellers
<10Hz
1- 25kB.
Advisory and modelling
The only AI solution working at the intersection of Control and Industry.
| Feature | PID | MPC | Cloud AI | Luffy ANC |
|---|---|---|---|---|
| Handles non-linear | ✗ | ✓ | ✓ | ✓ |
| Adapts to changes | ✗ | ✗ | ~ | ✓ |
| Runs on edge | ✓ | ~ | ✗ | ✓ |
| Easy to deploy | ✓ | ✗ | ✗ | ✓ |
- Only solution that adapts in real-time
- Runs on your existing hardware
- Handles non-linear systems
- Deploys in days, not months
