Key Metrics
2cm
New minimum detection range down from typical 7cm for D401/D405
10cm
New minimum detection range down from typical 30-50cm for D43x/45x series
Overview & live demo
The Improved Close Range Depth feature extends the RealSense usable depth range so the camera produces meaningful measurements much closer to the lens than the native hardware allows.
The update improves the minimum sensing distance (min-Z) down to about 2cm or 10cm depends on the model, past the native minimum, which typically sits around 30 cm and varies by camera model. It integrates transparently on top of the standard RealSense SDK pipeline. Available as a drop-in post-processing step or, on D500-series cameras, baked into the camera itself.
A side-by-side video showing raw hardware depth vs. the Improved Close Range Depth output is published in the upstream librealsense repository.
librealsense2 enhanced-depth-range README (v2.58.2):
github.com/realsenseai/librealsense/blob/r/2.58.2/examples/enhanced-depth-range/README.md
For the technical details Install steps, Python and C++ APIs, FrameMetadata, Calibration, and troubleshooting are documented in the upstream librealsense2 enhanced-depth-range README. This document stays high-level on purpose — the README is the source of truth.
Why close range matters
Most RealSense customers hit the native minimum (typically ~30 cm, varies by model) as a wall. Manipulation and bin-picking live below that line. So do humanoid hand-eye coordination, AMR navigation in tight spaces, retail interactions, and any cobot working at arm’s reach.
With Improved Close-Range Depth, depth becomes usable from roughly the width of your hand outward. Practical effects:
- AMRs & service robots — navigation through tight aisles, doorways, and around obstacles that pass within the native blind zone.
- Bin-picking — objects that previously fell into the blind zone now have valid depth.
- Hand-eye manipulation — the gripper’s working volume is in-range.
- Retail and kiosk — interactions at the counter no longer rely on RGB-only heuristics.
- Humanoids — wrist- or chest-mounted cameras see what the hands are doing.
D400 vs D500 deployment
The Improved Close Range Depth output working for all RealSense camera portfolio. For the D400, a library that runs on the host is required, while the D500 supports it natively on the SoC.
D400 SERIES · HOST-SIDE
Runs on a Jetson host.
The Improved Close Range Depth module runs in your host process, alongside the rest of the RealSense SDK pipeline.
Supports the installed D400 fleet D401/D405, D435i, D436, D455, D456, D457.
- No camera changes — works with cameras already shipping
- Available as a RealSense Viewer post-processing toggle
- Requires NVIDIA Jetson
D500 SERIES · ON-EDGE
Runs on the camera SoC.
D500 cameras carry the Gen5 SoC, so the Improved Close Range Depth enhancement runs on the camera itself. The host receives improved depth directly. no host-side compute, nothing extra to integrate.
- Zero host CPU / GPU cost
- Drop-in for existing depth-stream consumers
- Works on hosts without a GPU
Same output, two compute paths. A depth frame is a depth frame — improved close-range pixels look identical to your application whether they came from the host module on a D400 or directly from a D500 SoC.
Next steps
Improved Close Range Depth composes naturally with the rest of the Perception Studio catalog — and unlocks new behaviours for any robot working at arm’s length.
People Detection — on D500, Improved Close Range Depth and on-camera detection share the same SoC; both available without burdening the host.
Visual-Inertial Odometry (VIO) — pair localization with close-range manipulation depth on the same robot.
Try it. On D400 cameras (with a Jetson host), flip the Improved Close Range Depth toggle in RealSense Viewer to see it live. On D500 cameras, the improved depth is already on the depth stream — nothing to install.