RealSense unveils D585 Pro depth camera for robots

According to RealSense's press release and product page, RealSense unveiled the D585 Pro depth camera at Automate 2026. The vendor describes the device as an AI-native, software-defined vision platform built around a proprietary Gen 5 SoC that integrates a depth engine, image signal processor, DSP, dedicated AI accelerators and a quad-core Arm processor. RealSense product materials and coverage list the camera's headline specifications as a 120deg x 100deg field of view, depth output up to 1280 x 960, full-resolution imaging at 60 FPS, depth frame rates up to 90 FPS, and a minimum depth capability near 10 cm at full resolution with an ideal operating range beyond 10 m and a maximum exceeding 20 m. The Robot Report also reports the D585 Pro is expected to begin shipping in Q1 2027.

According to RealSense's product page and press materials, the D585 Pro uses dual infrared projectors, global-shutter imaging, integrated IR filters for indoor and outdoor operation, and IP65 protection as standard. RealSense states the camera delivers on-camera depth post-processing, an active noise-reduction compression mode that the company says reduces depth bandwidth by 75%, and an initial beta on-camera person detection feature. The company also announced a Perception Studio beta within the RealSense 2.0 SDK, with planned additions such as visual-inertial odometry, occupancy grid generation and face detection to be delivered to existing units via SDK updates.

Integrating a depth engine, ISP, DSP and AI accelerators on a single SoC follows a broader edge-AI trend where sensor vendors shift compute onto the sensor to cut host CPU/GPU load and reduce bandwidth. Companies building robot perception stacks typically balance trade-offs between sensor fidelity, host compute, and telemetry bandwidth; on-camera compression and on-device inference narrow those trade-offs by lowering the cost of high-resolution depth streams and enabling preliminary scene understanding at the sensor.

For robot integrators and perception engineers, the D585 Pro's combination of a wide 120deg x 100deg FOV, sub-15 cm (RealSense lists sub-15 cm and other materials cite 10 cm) close-range capability at full resolution, and multi-meter reach addresses a common need to cover manipulation workspaces and room-scale navigation with a single sensor. Equipping autonomous mobile robots (AMRs), humanoid platforms and inspection arms with a single sensor that handles both near-field manipulation and longer-range navigation can simplify sensor fusion and calibration work, provided the on-camera processing meets accuracy and latency requirements in real deployments.

Editorial analysis: Observers should track four indicators over the coming quarters:

• •adoption signals from robotics developers, which RealSense says are already evaluating the unit, per the press release and trade coverage;
• •the performance and maturity of Perception Studio and the SDK-delivered features, especially VIO and occupancy-grid outputs;
• •real-world bandwidth and latency measurements for the company's depth compression mode compared with host-side processing; and
• •shipping timing and pricing details, with The Robot Report noting an expected shipping start of Q1 2027.

If third-party benchmarks and integrator reports validate the claimed 2x-plus improvements in depth quality and close-range performance, the D585 Pro could shift how perception pipelines are architected in industrial and mobile robotics. Conversely, integration friction, driver support and edge-inference tooling will determine how quickly teams can replace multi-sensor stacks with a single, software-upgradeable camera.

For practitioners: monitor early integration notes, SDK release cadence and sample code from RealSense to evaluate whether the on-camera primitives (person detection, occupancy grids, VIO) match your latency and determinism needs before replacing existing sensors.