Kyber Brings VLC’s Low-Latency Roots to Robot Control

Kyber is turning a familiar software problem into a robotics infrastructure pitch: if machines are going to be operated from somewhere else, the video, sensor data, and control loop have to feel nearly local.

TechCrunch reported on June 19, 2026, that Jean-Baptiste Kempf, best known as a lead developer of VLC Media Player, has been building Kyber as an infrastructure layer for controlling remote devices in real time. The report says Kyber has raised a $5 million seed round led by Lightspeed and is targeting robotics, drones, remote IT access, and other systems where lag turns from annoyance into operational risk.

The news is not just another physical AI funding item. Kyber’s pitch is that remote machine control needs the same kind of low-level infrastructure discipline that made open-source video playback reliable at global scale. That puts the company in the middle of a practical question for robotics and AI operators: who owns the millisecond-sensitive layer between perception, command, telemetry, and human override?

A startup built around real-time control

Kyber’s own site describes the company as “real-time infrastructure for machine control,” spanning robots, drones, vehicles, remote desktops, and immersive environments. The company says its stack synchronizes video, audio, control inputs, and telemetry to a single clock with single-digit millisecond latency.

Video, sensors, and commands on one clock

That synchronization claim is the center of the story. A remote desktop session can tolerate some delay because a missed frame usually means a frustrated user. A robot arm, drone, vehicle, or remote industrial system is different: a delayed image or stale command can mean the operator is reacting to the past. Kyber’s public materials frame that gap bluntly, contrasting ordinary 200-millisecond remote experience with an 8-millisecond control target.

Those numbers should still be treated as vendor claims until independent deployments and benchmark conditions are public. But the architectural direction is credible: real-time machine control is less about a single video stream and more about keeping perception, control input, telemetry, encryption, and network adaptation aligned under jitter and packet loss.

The VLC connection matters

TechCrunch links Kyber’s approach to Kempf’s background in video-streaming technology. Kyber’s about page makes the same point more directly, saying the founding team built VLC, contributed to FFmpeg, and is extending years of video infrastructure and protocol work into machine-control domains.

Open source plus enterprise deployment

Kyber says the developer-facing project is open source and AGPL licensed, while enterprise customers can buy commercial licensing, support, and custom deployments. That hybrid model fits the market Kyber is chasing. Robotics and industrial operators often want transparency and sovereignty, but they also need someone accountable when latency, fleet updates, device support, or integration work breaks.

The company is also selling more than a library. TechCrunch reports that Kyber has about 25 full-time staff, offices in Paris, San Francisco, and Singapore, and a team that includes forward-deployed engineers for hands-on customer deployments. That makes the story closer to infrastructure operations than a simple SDK launch.

Why physical AI changes the risk model

The phrase “physical AI” can hide a lot of engineering detail. A model may decide what a robot should do, but the surrounding system still has to transmit video, collect sensor data, authorize tools, log behavior, handle degraded networks, and let a human take over when autonomy fails. Kyber’s site explicitly lists robots, drones, autonomous vehicles, remote desktops, immersive systems, and AI agents as target domains.

Human override becomes part of the platform

That is why low latency becomes a safety and governance issue, not just a performance feature. If an AI agent, autonomous vehicle, drone fleet, or warehouse robot needs human intervention, the fallback path has to be engineered before the incident. A remote operator needs a current view of the world, clear authority to act, and enough responsiveness to make the action meaningful.

In that sense, Kyber is betting that physical AI will need a control plane beside the model. The useful infrastructure will not merely stream pixels. It will make distance less visible to the operator while preserving enough security, observability, and auditability for enterprise environments.

Funding is a signal, not proof

Lightspeed’s public post says it led Kyber’s $5 million seed round as the company emerged from stealth, arguing that physical AI depends on the systems beneath it. The firm highlighted latency and precision as stakes rise for machines trusted across industries.

What buyers should watch next

The next proof points are not slogans about real time. They are named customers, deployment sizes, platform support, security reviews, network failure behavior, and independent measurements under messy conditions. Kyber claims protocol-level end-to-end encryption over QUIC/HTTP3 and broad platform support across Windows, Linux, Mac, Android, iOS, Quest, and Vision Pro. Operators will want to know how those claims hold up across fleet updates, constrained links, and mixed device classes.

Why this is worth watching

Robotics and physical AI are often discussed as model problems: better perception, better planning, better autonomy. Kyber’s launch is a reminder that deployment will also be an infrastructure problem. The system that keeps a human, an AI agent, and a machine in sync may decide whether a remote operation feels like control or just delayed observation.

If Kyber can turn VLC-era streaming lessons into a reliable control layer for robots, drones, and remote machines, it could become part of the connective tissue for physical AI. If not, the company will still be testing an important assumption: that the next wave of automation depends as much on protocol engineering and operations as it does on smarter models.

Sources

• TechCrunch: He made your free video player run smoothly. Now he’s doing that for robots.
• Kyber: Time for real-time
• Kyber: About us
• Lightspeed: Kyber raises $5M to improve physical AI infrastructure
• Featured image source: Robot arm handles an assay plate on Wikimedia Commons

Featured image: a laboratory robot arm handling an assay plate, courtesy of the National Institute of Allergy and Infectious Diseases; public domain licensing details are available on Wikimedia Commons. The image was resized and converted to WebP for sxz.io.