Shelly Robotics offers a cloud‑native robotics‑as‑a‑service platform that lets AI researchers upload models via API or SDK and execute them on a managed fleet of heterogeneous robots, with real‑time streaming of sensor data and performance metrics. The service handles hardware provisioning, safety constraints, environment resets, and provides consumption‑based or subscription billing, allowing teams to run perception and reinforcement‑learning experiments without maintaining physical robot labs.
Funding
Funding not disclosed
Founders
Product
Problem
Robotics research teams often face prohibitive costs and logistical overhead when provisioning physical robot hardware for AI model training and evaluation. Limited access to diverse platforms slows iteration cycles for perception, control, and reinforcement‑learning algorithms, especially for organizations without dedicated robot labs.
Solution
Shelly Robotics delivers a cloud‑native robotics‑as‑a‑service platform that abstracts away hardware procurement, maintenance, and scheduling. Users upload their AI models via a RESTful API or SDK, and the system dispatches them to a managed fleet of heterogeneous robot units for real‑world execution. The platform streams sensor data, control commands, and performance metrics back to the user’s workspace in real time, enabling closed‑loop training and rapid hypothesis testing. Built‑in safety envelopes and automated environment resets ensure repeatable experiments without manual intervention. Billing is consumption‑based or via tiered subscriptions, allowing teams to scale compute‑intensive robotics workloads on demand.
Target Audience
Primary customers are AI robotics research groups in academia, corporate R&D labs, and early‑stage startups that need on‑demand access to physical robots for model validation and iterative development.
Features
- Multi‑robot fleet orchestration with automated provisioning, health monitoring, and remote firmware updates
- SDKs for Python, ROS 2, and TensorFlow/PyTorch that expose robot APIs for perception, actuation, and reinforcement‑learning loops
- Real‑time bidirectional streaming of high‑resolution camera, LiDAR, IMU, and joint‑state data over secure WebSocket channels
- Containerized execution sandbox that isolates user code, enforces safety constraints, and auto‑resets robot environments after each run
- Scalable job scheduler that batches experiments across heterogeneous platforms (mobile bases, manipulators, quadrupeds) to maximize utilization
- Integrated data lake with versioned experiment logs, metadata tagging, and export to cloud storage for downstream analysis
- Role‑based access control and end‑to‑end TLS encryption complying with ISO 27001 security standards
- Dashboard and CLI for monitoring job status, visualizing telemetry, and retrieving deterministic replay recordings