Leopard Imaging and Lumotive say their Sirius Max camera measures depth beyond 20 meters, indoors and out, from a single GMSL2 module that also carries a global-shutter RGB sensor. Both companies are in Fremont, CA. The claim is a vendor figure; no third-party test conditions were published. But the architecture underneath is worth an engineer's attention.

Sirius Max merges three functions that robots today usually get from separate boxes: RGB imaging, indirect time-of-flight (iToF, which infers distance from the phase shift of returned modulated light), and programmable beam-steered illumination. Synchronized RGB, depth, and control all run over one GMSL2 coax cable (a high-speed serial link for automotive-grade cameras). One module replaces several sensors. That's the whole pitch.

Why steer the light at all?

Flash iToF, the incumbent approach, floods the entire field of view with laser power. Very little light returns from distant objects, so range suffers, and light bouncing off walls and floors corrupts depth readings near corners. Dr. Apurva Jain, senior vice president of product, marketing, and customer success at Lumotive, laid out the physics: "Flash iToF spreads the laser power over the entire field of view, so very little light returns from distant objects."

The Sirius Max instead illuminates discrete regions of interest and stitches them into a full point cloud, concentrating more optical power per region. That higher signal-to-noise ratio is what extends range past 20 meters while keeping short-range accuracy, according to the company. The steering comes from Lumotive's Light Control Metasurface (LCM), a solid-state semiconductor chip that shapes and directs laser light electronically. No moving parts. Mechanical LiDAR, by comparison, adds cost, size, and failure-prone mechanics.

What can developers actually program?

Steering angle, beam divergence, dwell time, and scan pattern are all software-configurable in real time, with task presets such as an AMR mode. A robot navigating an aisle can scan the far field for obstacles while dedicating dense coverage to the floor ahead for cliff detection. During manipulation, it can concentrate resolution on a close-range region of interest.

The module also handles reflective and high-contrast surfaces with on-the-fly HDR: it identifies saturated pixels and re-exposes them within the same frame, so dark and bright objects both survive. The depth sensor is the onsemi AF0130 iToF device. Jain's framing: "Programmable optics lets light be controlled as flexibly as software, so the sensor becomes a programmable eye that the AI stack can direct to the information it needs." One platform, then, can serve SLAM (simultaneous localization and mapping), collision avoidance, cliff detection, inspection, and navigation.

Measured performance versus claims

Ambient light rejection and multipath (spurious depth errors from light bouncing off multiple surfaces) improve because directional illumination concentrates the signal and lights one region at a time rather than the whole scene. Jain said: "In side-by-side testing against flash iToF and active stereo, LCM-based iToF produces point clouds with less noise, fewer artifacts and blind spots, better near-range precision and more reliable detection of people, pallets, floors and corners."

Note the test description: side-by-side, conditions unspecified. The 20-meter range figure and the comparative point-cloud claims come from the vendors, not an independent lab. Buyers with bright facilities or shiny conveyor belts will want to run their own corner-case tests. Steering also reduces interference between neighboring units, which matters for multi-robot fleets.

Who benefits, and does the ROI hold?

The primary beneficiaries are warehouse and factory AMRs and AGVs, where longer-range vision enables safer high-speed operation, and humanoid robots, which need close-range precision for manipulation plus longer range for navigation from one compact sensor. Outdoor robotics and smart infrastructure round out the list.

On cost, Lumotive points to deployments with major OEMs, claiming significantly reduced total cost of ownership:

• One sensor instead of two or three
• Solid-state reliability, less maintenance, more uptime
• Lower power, less heat, longer battery life
• Higher frame rates for greater throughput
• Better precision, so robots work closer to walls and people with fewer false stops

No customer names, no numbers attached. Treat the ROI list as directional.

The adoption question

The consolidation argument is sound where space and battery budgets are tight — humanoids, drones, compact AMRs. The open question is integration effort. A software-defined sensor is only as good as its SDK, and fleets that tune their sensing in real time need the tooling to prove that tuning works. Engineers specifying perception for 2025-26 platforms should ask two things: what do the point clouds look like in my worst-case lighting, and who owns the beam-steering firmware when the AI stack wants a new scan pattern mid-shift?