Guild Robotics
Guild Robotics develops AI-driven robotics systems for logistics, deploying quadruped, forklift, and inventory robots that operate autonomously in warehouses and commerce environments. The company runs its π0.5 vision-language-action model on edge hardware at 43 Hz using advanced distillation, compilation, and quantization techniques to enable real-time decision-making without cloud dependency.
- Artificial Intelligence
- AI Agents
- Hardware
- Industrial Automation
- Logistics & Supply Chain
- Robotics
- Software Only
Funding
Founders
Product
Problem
Traditional logistics operations rely on manual material handling, which is labor-intensive, error-prone, and difficult to scale. Existing automation solutions often require extensive environment modifications, struggle to adapt to dynamic warehouse layouts, and depend on cloud connectivity, which introduces latency and reliability risks for real-time robotic control.
Solution
Guild Robotics builds autonomous robots for material movement across logistics environments, including quadruped platforms, inventory robots, forklifts, and robotic arms. The company focuses on deployment simplicity—robots adapt to existing environments without infrastructure changes—and edge intelligence, running their π0.5 vision-language-action model locally on Jetson Thor hardware at 43 Hz. This approach enables real-time perception and decision-making with minimal latency, allowing robots to navigate, manipulate goods, and complete tasks safely alongside human workers in warehouses and distribution centers.
Target Audience
Primary customers are warehouse operators, e-commerce fulfillment centers, and third-party logistics providers that need scalable, autonomous material-handling solutions in dynamic indoor environments.
Features
- Quadruped, inventory, forklift, and arm robot form factors covering diverse material-handling tasks
- π0.5 VLA model deployed on edge hardware at 43 Hz inference speed for low-latency autonomous control
- Shallow-π knowledge distillation compresses the 2.6B parameter model by subsampling layers, halving latency from 260ms to 127ms
- SnapFlow single-step denoising replaces 10-step diffusion, reducing latency 2.3x to 56ms
- TensorRT compilation with CUDA engine optimization and ONNX graph control achieves 1.3x speedup to 45ms
- FP8 quantization via NVIDIA ModelOpt preserves accuracy by calibrating on task-specific datasets like LIBERO and DROID
- Environment-adaptive design requires no fixed infrastructure or environment modifications