PICO and NVIDIA Jointly Launch Isaac Teleop, an Open-Source ImmersiveRobot Teleoperation Solution

At the recent NVIDIA GTC 2026 conference, PICO and NVIDIA jointly launched

Isaac Teleop, an open-source immersive robot teleoperation solution. It is not a single-

point product, but a standardized framework designed to address the underlying data

challenges in the robotics industry.

Although the concept of Physical AI is drawing strong attention, the industry still

widely faces practical problems such as high training-data costs, severe system

fragmentation, and a disconnect between simulation and real-world environments.

PICO and NVIDIA aim to solve this by using XR technology to create an operational

loop between humans and robots, turning data collection from specialized lab

equipment into a standard tool for developers.

Isaac Teleop uses a four-layer architecture. It provides a unified device interface,

enabling precise 6-DoF control with the PICO 4 Ultra, controllers, and motion

trackers. It also offers a unified runtime environment, allowing simulation code from

Isaac Lab to be used directly for controlling real robots through Isaac ROS. In

addition, it uses the first-person video recording function of PICO headsets to collect

training data, and the upcoming Ego4Robo feature will further simplify the process.

Cloud deployment is supported through Volcano Engine, which reduces enterprise

hardware barriers.

PICO already has experience in robotics. Its XRoboToolkit has served more than 100

robotics organizations and won the Best Paper Award at IEEE SII 2026. Through this

cooperation with NVIDIA, teleoperation is being expanded from a local tool to a

cloud-based ecosystem. The next-generation headset Project Swan will also be deeply

integrated with Isaac Teleop. The solution already supports PICO 4 Ultra and WebXR

access, and the toolchain is now open-source.

In the short term, it lowers the barrier to data collection. In the medium term, its

standardized format may enable data interoperability. In the long term, if Ego4Robo

matures, human operation videos may be converted directly into robot training data.

However, data bias caused by operator skill differences and latency in cloud

deployment still need to be verified in real applications. (Source: VRAR WORLD)