Launch HN: OctaPulse (YC W26) – Robotics and computer vision for fish farming

Introduction to OctaPulse: Revolutionizing Fish Farming with Robotics and Computer Vision

As a tech enthusiast, I'm always excited to come across innovative solutions that address real-world problems. Recently, I stumbled upon OctaPulse, a startup that's making waves in the fish farming industry with its cutting-edge robotics and computer vision technology. In this post, we'll delve into the world of OctaPulse and explore how they're transforming the way fish are farmed.

The Problem: Inefficient Fish Farming Practices

Fish farming is a $350 billion global industry, yet it still relies on labor-intensive and inefficient practices. For instance, farmers need to measure their fish stock regularly to make informed decisions about feeding, breeding, and harvesting. However, this process is often done manually, which can be time-consuming and stressful for the fish. OctaPulse is on a mission to change this by introducing automation and data-driven insights to the industry.

How OctaPulse Works

OctaPulse uses a combination of robotics and computer vision to automate fish inspection and measurement. Their system consists of:

• Luxonis OAK cameras, which provide depth and RGB data in a compact form factor
• Nvidia Jetsons for heavier workloads like segmentation and keypoint extraction
• Custom-built robotics enclosures with soft robotics grippers for handling fish
• A proprietary software platform for labeling, task assignment, and model management

The OctaPulse system uses YOLO variants for detection, custom segmentation heads for body outlines, and keypoint models for anatomical landmarks. However, getting these models to run efficiently on edge hardware has been a challenge. The team has had to experiment with TensorRT, OpenVINO, and ONNX Runtime to achieve the required speed and accuracy.

Overcoming Challenges

One of the biggest challenges OctaPulse faced was dealing with the harsh environment of fish farms. Saltwater, humidity, and water turbidity all posed significant technical challenges. To overcome these, the team had to develop custom solutions, such as:

• Building calibration datasets that capture the variance in lighting, water clarity, and fish density
• Using INT8 quantization on TensorRT to achieve the required speed, while carefully monitoring accuracy degradation
• Designing compliant grippers that can safely handle fish without damaging them

The Future of Fish Farming

OctaPulse is not just about automating fish inspection; it's about transforming the entire fish farming industry. By providing accurate and reliable data, farmers can make informed decisions about breeding, feeding, and harvesting. This can lead to:

• Improved fish yields and quality
• Reduced waste and environmental impact
• Increased efficiency and profitability for farmers

Who is this for?

OctaPulse is perfect for:

• Fish farmers looking to improve their operations and increase efficiency
• Researchers and scientists interested in aquaculture and computer vision
• Anyone passionate about sustainable food production and reducing environmental impact

As I read through the comments on the original post, I was struck by the enthusiasm and interest in OctaPulse's technology. The team is eager to hear from experts in computer vision, edge deployment, and aquaculture, and I'm excited to see how their solution will evolve.

What do you think about OctaPulse's innovative approach to fish farming? Do you have any experience with computer vision or robotics in harsh environments? Share your thoughts and questions in the comments below!