Sairlab develops algorithms and systems that give mobile robots human‑level autonomy by combining perception, spatial reasoning, and decision‑making. Their core approach uses a neuro‑symbolic learning framework that integrates neural and symbolic memory for data‑efficient, self‑supervised learning, enabling robots to perceive images, point clouds, and proprioceptive data, and to plan and act in real time in unstructured environments. Sairlab also maintains PyPose, an open‑source differentiable robotics library that has surpassed 260,000 downloads across 2025‑2026.
Funding
Funding not disclosed
Founders
Product
Problem
Mobile robots operating in unstructured, dynamic environments struggle to reliably interpret heterogeneous sensor data and to plan actions in real time without extensive labeled datasets. Existing approaches often separate perception, reasoning, and control, leading to brittle performance and limited adaptability.
Solution
Sairlab develops a unified neuro‑symbolic framework that fuses visual, point‑cloud, and proprioceptive inputs with neural and symbolic memory to provide spatial common sense and semantic understanding. The system enables data‑efficient, self‑supervised learning, allowing robots to acquire useful representations without large labeled datasets. Real‑time planning and decision‑making are achieved through differentiable algorithms that operate on manifold representations of robot states. An open‑source library, PyPose, supplies modular, differentiable robotics tools that support end‑to‑end training and deployment on mobile platforms.
Target Audience
Primary customers are robotics research labs, autonomous vehicle developers, and industrial automation teams that require advanced perception‑reasoning‑action capabilities for mobile robots operating in complex, real‑world settings.
Features
- Integrated perception pipeline that processes images, 3D point clouds, and proprioceptive signals within a single neuro‑symbolic model
- Neural‑symbolic memory structures that encode spatial relationships and semantic knowledge for robust reasoning
- Self‑supervised learning mechanisms that reduce dependence on annotated data and improve data efficiency
- Real‑time, differentiable planning algorithms operating on manifold representations of robot pose and dynamics
- Open‑source PyPose library offering differentiable robotics primitives, manifold optimization, and modular components for research and deployment
- End‑to‑end trainable architecture that combines perception, reasoning, and control in a unified framework