What this post covers:
- The engineering behind Improved Close-Range Depth
- Design trade-offs
- Supported cameras
- Performance improvements
- When to use it
Improved Close-Range Depth: Extending RealSense Stereo Depth with Min-Z
How Improved Close-Range Depth extends sensing to approximately 2 cm on D401/D405 and 12 cm across D43x/D45x cameras.
Robots often need their most accurate spatial information at the point where conventional stereo depth becomes most difficult to use close to the camera.
A robotic gripper approaches an object. A humanoid reaches toward a surface. An inspection camera moves within centimeters of a component. An autonomous mobile robot passes close to a shelf, doorway, or unexpected obstacle.
In each case, the system needs reliable depth inside the camera’s immediate working area. Yet stereo depth cameras have a native minimum sensing distance. When an object moves inside that range, disparity becomes increasingly difficult to resolve, and the resulting depth data may become incomplete or unusable.
Improved Close-Range Depth—also known as Min-Z—is designed to extend usable stereo depth into this near-field region.
Available through RealSense Perception Studio, Min-Z extends sensing to approximately 2 cm on the RealSense D401 and D405 and approximately 12 cm across supported D43x and D45x cameras. It gives developers a way to bring valid depth closer to the lens without replacing compatible cameras or creating a separate sensing pipeline for near-field tasks.
Why Stereo Depth Has a Minimum Range
Stereo depth works by comparing images captured from two slightly different viewpoints. The system identifies corresponding features in the left and right images and measures their horizontal displacement, or disparity.
The relationship is conceptually straightforward: objects with greater disparity are closer to the camera, while objects with smaller disparity are farther away.
At very close distances, however, several constraints begin to converge.
The disparity required to represent the object may exceed the range supported by the native stereo configuration. Parts of the scene may appear in one imager but not the other. Occlusion increases. Matching becomes more difficult, particularly around edges, reflective materials, repetitive textures, and surfaces with limited visual detail.
The result is a near-field region where the camera may still see the object in its individual images but cannot produce sufficiently complete and reliable depth through the standard pipeline.
This is the problem Min-Z addresses.
Extending the Usable Depth Range
Improved Close-Range Depth operates as an enhancement to the existing RealSense depth pipeline. Rather than requiring developers to replace their application’s depth interface, it produces an enhanced depth frame that can be consumed by downstream applications in the same way as standard depth data.
The objective is not simply to generate more pixels close to the camera. The objective is to recover useful spatial information in areas where the native stereo output would otherwise contain gaps or invalid depth.
The exact minimum range depends on the camera architecture:
- D401 and D405: approximately 2 cm
- D43x and D45x cameras: approximately 12 cm
- D500 series: enhanced close-range processing can run natively on the camera
Figure 1. Side-by-side comparison of raw stereo depth and Min-Z improved output. Min-Z recovers usable depth in close-range regions that are missing from the native depth stream.
These distances should be understood as practical operating targets rather than universal guarantees. Performance can vary with the camera model, scene geometry, lighting, object material, texture, exposure settings, and required accuracy.
That distinction matters. Min-Z extends the useful operating envelope of stereo depth, but it does not eliminate the physical and optical trade-offs of close-range sensing.
Two Deployment Paths
Perception Studio supports different processing paths based on the capabilities of the camera platform.
Figure 2. D400 cameras use host-side Min-Z processing on NVIDIA Jetson, while D500 cameras perform the enhancement directly on the camera SoC. Both paths deliver enhanced depth to the robot application.
D400 Series: Host-Side Processing
For supported D400-series cameras, Improved Close-Range Depth runs as part of the host processing pipeline on an NVIDIA Jetson platform.
This approach allows developers to evaluate the capability using compatible RealSense cameras already deployed or in development. The enhancement can be added without modifying the camera hardware or changing the fundamental depth-stream interface used by the application.
The trade-off is that the host must provide the compute required for the enhanced processing. Teams should account for that workload alongside navigation, planning, inference, and other robotic functions already running on the Jetson platform.
D500 Series: On-Camera Processing
On D500-series cameras, the capability can run on the camera’s Gen 5 system-on-chip.
The host receives the enhanced depth output directly, without needing to execute the close-range processing itself. This reduces the impact on host CPU and GPU resources and simplifies integration for systems in which compute is already tightly allocated.
For the application, the underlying goal is the same: provide an improved depth stream with more usable information inside the native near-field region. The difference is where that processing takes place.
D500 Series: On The Engineering Trade-Offs – Camera Processing
Extending stereo depth closer to the camera involves trade-offs. Developers should evaluate Min-Z against the requirements of the complete application rather than treating minimum distance as the only performance metric.
Close-Range Coverage Versus Native Measurement Conditions
Min-Z is most valuable where standard depth contains missing or invalid regions. It expands coverage, but recovered depth should still be evaluated for the accuracy, stability, and edge quality required by the application.
A presence-detection system may tolerate different measurement characteristics than a precision inspection or grasp-planning system.
Scene and Material Dependence
Close-range stereo performance remains sensitive to the scene.
Low-texture surfaces can make correspondence difficult. Reflective, transparent, or highly absorptive materials may produce unstable results. Strong lighting, occlusion, and extreme object angles can also affect the available depth.
Min-Z improves the operating range, but application testing should include representative objects, lighting, motion, and mounting geometry.
Host Compute Versus On-Camera Processing
On D400-series deployments, enhanced processing consumes resources on the NVIDIA Jetson host. This can be a practical trade-off when extending the capabilities of an existing camera fleet, but the workload should be measured as part of the total system budget.
D500-series on-camera execution avoids that host-side burden and may be preferable when compute headroom, latency, or architectural simplicity is a priority.
Near-Field Performance Versus Full-Range Optimization
A camera configuration optimized for close manipulation may not be identical to one optimized for maximum range or a broad navigation environment.
Robots that must operate across both near- and far-field conditions should validate transitions between those regions and determine whether one configuration satisfies the entire task or whether application-level modes are appropriate.
What Changes for the Developer
From the application’s perspective, the key benefit is continuity.
Improved Close-Range Depth integrates with the RealSense SDK pipeline rather than requiring a separate close-range sensor interface. Existing depth consumers can continue working with depth frames while gaining access to enhanced near-field information.
This can reduce the amount of custom logic developers need to create around the native minimum range. Instead of treating the area near the camera as an unavoidable blind zone, the application can continue using depth deeper into the robot’s working envelope.
That can simplify:
- Grasp and approach planning
- Hand-eye coordination
- Collision and proximity checks
- Close-up measurement
- Object segmentation near the camera
- Navigation around nearby obstacles
The technical tutorial accompanying the Perception Studio release provides the implementation steps, supported packages, APIs, and configuration guidance. The purpose of this article is to explain where Min-Z fits architecturally and what teams should consider when evaluating it.
Where Min-Z Creates the Most Value
Manipulation and Bin Picking
Objects frequently enter the native minimum range as a robot arm or gripper approaches them. Losing depth during the final stage of an interaction can force developers to rely on prior estimates, RGB-only processing, or additional sensors.
Min-Z helps preserve depth information deeper into the approach, supporting grasp refinement, object localization, and collision avoidance.
Figure 3. Min-Z extends usable depth into the robot’s immediate working area, where accurate perception is essential for approaching, locating, and grasping objects.
Humanoid Hand-Eye Coordination
Humanoid cameras may be mounted on the head, chest, or wrist. In each configuration, the hands and manipulated objects can move close to the sensor.
Improved close-range depth can give the perception system better visibility into the area where physical interaction occurs, particularly for reaching, handoffs, and tool use.
Inspection and Measurement
Inspection systems often move close to a component to detect small features or analyze surface geometry.
Min-Z expands the range in which depth can contribute to dimensional analysis, defect inspection, robotic alignment, and scan planning. Applications with strict metrology requirements should validate accuracy against their specific tolerances.
Tight-Space Navigation
AMRs and service robots can encounter nearby obstacles that enter the native depth blind zone, including shelving, doorframes, furniture, and objects protruding into the path.
Improved near-field coverage can complement the broader navigation stack by supplying depth closer to the camera, where last-moment proximity awareness may be important.
Retail and Human Interaction
Robots, kiosks, and interactive systems often operate across counters or at arm’s length. At those distances, close-range depth can support gesture-aware interaction, object exchange, occupancy detection, and spatial understanding without relying exclusively on RGB inference.
When Min-Z May Not Be the Right Choice
Min-Z is most useful when the application’s critical working volume extends inside the camera’s native minimum sensing range.
It may add limited value when:
- All relevant objects remain comfortably beyond the native Min-Z
- The system has no available Jetson compute for a D400 host-side deployment
- The application requires measurement tolerances that have not been validated in the enhanced region
- Reflective, transparent, or textureless materials dominate the use case
- A purpose-built short-range sensor already satisfies the requirement
The right evaluation question is not simply, “How close can the camera see?” It is:
Does the enhanced depth provide sufficiently complete, stable, and accurate information in the working volume that matters to the robot?
How to Evaluate Min-Z
A meaningful test should reproduce the conditions of the intended application.
Developers should evaluate:
- Minimum and maximum operating distance
- Target object materials and textures
- Lighting conditions
- Camera and object motion
- Occlusion around tools or grippers
- Depth completeness and temporal stability
- Edge behavior
- Host resource use on D400 deployments
- Accuracy at task-critical points
- Transitions between enhanced close range and the standard depth region
A side-by-side comparison between the standard and enhanced depth outputs can quickly show where Min-Z increases coverage. Task-level testing is then needed to determine whether that improvement translates into more reliable robotic behavior.
Extending the Visual Cortex of Physical AI™
Robotic perception must work across the full physical interaction space—from broader environmental awareness to the final centimeters around a gripper, hand, or obstacle.
Improved Close-Range Depth extends the RealSense perception stack into an area where stereo systems have traditionally faced limitations. It is one example of how Perception Studio expands RealSense’s role as the Visual Cortex of Physical AI™ through continuously evolving software and edge-processing capabilities.
For developers, the value is practical: greater near-field visibility, a consistent depth pipeline, and a faster path to testing close-range behaviors on compatible RealSense hardware.
Get Started
Improved Close-Range Depth is available through the Perception Studio early-access program.
The accompanying tutorial includes installation guidance, supported camera configurations, software requirements, API examples, and troubleshooting information.
Explore the Min-Z tutorial and test Improved Close-Range Depth in your application.