Verapulse provides efficient, on‑device, real‑time control intelligence for robots, running on existing hardware such as Jetson modules or bare CPUs without requiring a GPU or cloud connection. Their flagship offering is an open‑source, compact Vision‑Language‑Action (VLA) model that can be accessed via a simple hosted API and delivers top performance on the LIBERO benchmark, enabling cost‑effective, low‑latency deployment at fleet scale.
Funding
Funding not disclosed
Founders
Product
Problem
Robotic systems often require powerful GPUs or cloud-based inference to run vision-language-action models, leading to high hardware costs, latency from network roundtrips, and limited scalability for large fleets.
Solution
Verapulse provides an open-source, compact Vision‑Language‑Action (VLA) model that can run in real time on existing robot hardware such as NVIDIA Jetson modules or even bare CPUs. The model is distributed via a simple hosted API, allowing developers to integrate on‑device intelligence without adding per‑robot GPUs or relying on cloud services. Benchmarked as the top performer in its class on the LIBERO suite, the solution delivers an order‑of‑magnitude reduction in both cost and latency at fleet scale, making advanced robot control feasible on the hardware that most real‑world robots already use.
Target Audience
Primary customers are robotics companies and research teams that need real‑time, cost‑effective control for fleets of robots operating on standard edge hardware.
Features
- Fully open‑source VLA model optimized for low‑power edge processors (Jetson, CPU)
- Hosted API for easy integration into existing robot software stacks
- State‑of‑the‑art performance on the LIBERO benchmark, outperforming comparable models
- No per‑robot GPU requirement, eliminating expensive hardware upgrades
- On‑device inference eliminates cloud round‑trip latency, improving reliability
- Designed for deployment across large robot fleets with minimal cost impact