Why Video Processing Matters for the Next Generation of Robots

Mobilint has begun mass production of next-generation edge AI chips that incorporate video codec technology from Chips&Media, the companies announced on November 4. The South Korean partners are targeting robotics applications, including humanoids, where several cameras can generate more video than a system can easily store or transmit. The production milestone moves a longstanding licensing relationship into manufacturing and gives robot developers another option for handling video alongside on-device AI.

The partnership began with licenses for H.264/AVC, H.265/HEVC and AV1. Mobilint subsequently added licenses for AV2 and H.266/VVC, broadening the formats available for future designs. At the center of the announcement is Chips&Media’s dedicated video processing unit, which the company says can encode multiple streams and handle 8K video at 60 frames per second on a single core. The release leaves open which formats the production chips support at that throughput and how the hardware performs across different combinations of camera feeds.

A robot may use forward-facing cameras to navigate and wrist-mounted cameras to see what it is grasping. It may also need to record an incident, send a live feed to a remote operator or retain footage for debugging. Chips&Media estimates that a robot chip can be called on to handle eight to 20 camera channels simultaneously. Those feeds create distinct workloads. A navigation model needs timely images, while a recording system needs to compress video efficiently enough to store it.

Dedicated encoding hardware can take some of that work off the general-purpose processor. In a battery-powered robot, that can help engineers manage the power and heat associated with keeping several feeds active. It can also make it easier to capture footage for diagnostics while the robot continues working. The practical test is how many streams the chip can handle at the resolutions and frame rates a robot actually uses, within its power budget and while its AI models are running. The headline 8K figure gives designers a starting point for that evaluation.

Functional safety is another reason Chips&Media says Mobilint selected its technology. The supplier’s video codec IP has ISO 26262 ASIL-B certification, an automotive functional safety credential. For a robotics team, evidence about how a component detects and handles faults can help with integration. The robot maker still has to evaluate the full system, including what happens if a camera feed freezes or arrives late and how the machine responds when it can no longer rely on that input.

The companies also plan to explore frame buffer compression and data compression. Frame buffer compression can reduce the amount of image data that must move between processing hardware and memory. That matters when camera feeds, video encoding and AI inference all compete for memory bandwidth. If the collaboration expands into compressing intermediate AI data, it could address another source of memory traffic inside the chip. The companies describe that broader work as a plan.

Bottom line: For Chips&Media, the deal puts video technology developed for markets including consumer devices and automotive systems into a chip aimed at physical AI. For Mobilint, it adds a dedicated video subsystem to its pitch to robotics manufacturers. The announcement leaves production volumes, named robot customers and deployment schedules unspecified. The next evidence of commercial traction will come from manufacturers choosing the chips for specific robots and showing how those systems perform under sustained camera and inference workloads. For a factory robot, that could mean recording several views through a full shift. For a remotely supervised machine, it could mean keeping an operator’s video feed responsive while the robot processes its surroundings locally. Those are the kinds of results that will show what the integration buys robot makers in practice.

Contact Chips&Media

Also Read:

Chips&Media targets AI memory bottlenecks with new compression IP

Fail-Safe Frames: Why Chips&Media’s ASIL-B Video IP Matters

WAVE-J: Redefining High-Performance JPEG for the 8K Era