Founded by 24-year old IIT Madras graduates Sreeraj R and Rishabh Sharma, Panoculon Labs builds specialised cameras for collecting first-person, or egocentric, data used by robotics companies and data vendors to train physical AI systems.
Today, physical AI is facilitating machines such as robots, drones, and automated vehicles to understand the real world so that they can take action on their own.
However, in order to make physical AI effective and useful, they require large volumes of high-quality real-world data. This is where the industry faces a major challenge.
Tech and industrial organisations are collecting data with generic consumer devices such as GoPros and iPhones which are not specifically designed for this purpose. First, these devices are costly. GoPros can cost around Rs 20,000-25,000 while a good quality latest iPhone model can cost around Rs 50,000-90,000.
Secondly, these devices weigh around 200 gms or more which can make a person uncomfortable when worn for an 8-9 hour shift, causing headaches and strain.
Also, these devices are not suitable for training a good model because robotics labs can also have specific requirements around video compression, including H.264/H.265 encoding and GOP frame ranges, which consumer devices are not specifically designed to address.
This is the gap that Bengaluru-based Panoculon Labs is solving.
“GoPros and iPhones are very costly consumer devices. More importantly, they are not designed specifically for data collection. For physical AI, there are very specific requirements around field of view, sensor synchronisation, recording formats and other technical parameters,” Rishabh Sharma tells Startup Pedia in an exclusive interview.
The startup designed a device Trinet, which is a lightweight egocentric camera designed specifically for robotics data collection. The device can be worn on the wrist.
The device weighs 40 grams, offers a 150-degree horizontal field of view and combines video with a 9-DoF IMU operating at 400Hz.
Trinet can stream video and IMU data through a standard UVC interface or record directly onto a microSD card. It also supports tightly synchronised multi-camera data collection for head and wrist cameras.
The company says its hardware-level synchronisation keeps camera and motion data on a shared timeline, addressing a key requirement for robotics datasets.
Panoculon has also developed Trinet Stereo, which uses two RGB cameras, a 70mm baseline, a single-clock trigger and an integrated 9-DoF IMU for depth capture.
The startup has sold around 1,500+ units in B2B mode since starting its main production line, with customers including robotics companies and data vendors.