Skip to main content
📸 Sensor GuideUpdated June 2026 · 6 cameras tested

Best Depth Cameras for Robots 2026

Intel RealSense, Stereolabs ZED, Luxonis OAK-D, and Orbbec head-to-head. Six depth cameras ranked by range, accuracy, ROS 2 support, GPU requirements, and robotics use-case fit.

🏆 #1 ZED 2i — Best Outdoor Range🥇 #2 RealSense D455 — Best All-Rounder⚡ #4 OAK-D Pro W — Best On-Device AI
#1
Best Outdoor RangeStereo

ZED 2i

Stereolabs · $449

93
Range
0.2 – 20 m
Accuracy
< 1% at 5 m
FoV (H)
110° horizontal
Weight
166 g
IMU
Yes ✓
ROS 2
Official
GPU
CUDA-required
Resolution
4416
Outdoor NavDroneIndoor SLAM
✓20 m depth range — best for outdoor mobile robots and drones
✓IP66 rated — dustproof, waterproof for field robots
✓Built-in IMU fused with depth for VIO (visual-inertial odometry)
✗Requires CUDA GPU (Jetson Orin or desktop RTX) — no CPU-only mode
✗$449 — most expensive in this comparison
✗Needs USB 3.0 bandwidth — can cause issues on embedded boards

The ZED 2i is the gold standard for outdoor robotics and drones where range, robustness, and VIO accuracy matter. If your robot operates outdoors, choose this. For pure indoor use, the $300 price premium over RealSense is hard to justify.

#2
Best All-RounderStereo

Intel RealSense D455

Intel · $299

91
Range
0.4 – 6 m
Accuracy
< 2% at 4 m
FoV (H)
87° horizontal
Weight
120 g
IMU
Yes ✓
ROS 2
librealsense2
GPU
CPU-only
Resolution
1280
Indoor SLAMManipulationEducation
✓No GPU required — runs on Raspberry Pi, Jetson Nano, any x86 machine
✓Excellent ROS 2 ecosystem — 1000s of tutorials, Nav2 integration examples
✓Global shutter on D455 — less motion blur for moving robots
✗6 m max range — not suitable for large outdoor spaces
✗D455 uses passive IR — struggles in direct sunlight
✗Depth quality on fast-moving robots degrades (use D435i for fast motion)

The D455 is the most practical choice for indoor robots, research arms, and educational projects. No GPU dependency, strong ROS 2 support, and the widest tutorial base of any depth camera. Start here unless you need outdoor range.

#3
Best Budget StereoStereo

Intel RealSense D435i

Intel · $199

87
Range
0.1 – 10 m (typ. 3 m reliable)
Accuracy
< 2% at 2 m
FoV (H)
87° horizontal
Weight
72 g
IMU
Yes ✓
ROS 2
librealsense2
GPU
CPU-only
Resolution
1280
Indoor SLAMManipulationEducationDrone
✓$199 — most affordable stereo depth camera with IMU
✓72 g — light enough for drone integration
✓Active stereo IR illumination — works in dark environments
✗Rolling shutter causes blur on fast robots or drones
✗3 m reliable depth range (10 m spec is optimistic)
✗Discontinued/limited stock in some regions — D455 is successor

The D435i remains the best budget-friendly stereo camera for indoor robots and lightweight drones. If your robot moves slowly and stays indoors, the $100 savings vs D455 is worth it. For fast-moving platforms, step up to the D455 or ZED.

#4
Best On-Device AIStereo

Luxonis OAK-D Pro W

Luxonis · $349

85
Range
0.2 – 12 m
Accuracy
< 2% at 4 m
FoV (H)
120° horizontal (wide)
Weight
91 g
IMU
Yes ✓
ROS 2
depthai-ros
GPU
On-device
Resolution
4056
ManipulationWarehouseDroneIndoor SLAM
✓On-device Myriad X VPU — run YOLOv8 at 30fps without a host GPU
✓120° wide-angle for manipulation robots needing broad coverage
✓DepthAI SDK simplifies AI pipeline deployment significantly
✗$349 — not cheap for a camera without external GPU
✗ROS 2 wrapper is less mature than RealSense's
✗Myriad X is limited to ~4 TOPS — large models need offloading

OAK-D Pro W is uniquely suited for robots that need edge AI inference on the camera itself — think sorting robots, pick-and-place systems, and drones that can't afford host GPU latency. If on-device inference is your priority, no other camera at this price beats it.

#5
Best ToF AccuracyToF

Orbbec Femto Mega

Orbbec · $999

81
Range
0.25 – 5.5 m
Accuracy
< 5 mm at 1 m
FoV (H)
120° horizontal
Weight
269 g
IMU
Yes ✓
ROS 2
OrbbecSDK
GPU
CPU-only
Resolution
1024
ManipulationWarehouseIndoor SLAM
✓iTOF with sub-5mm accuracy at close range — best for precise manipulation
✓1M pixel ToF — highest resolution in this list
✓120° FoV on ToF (most ToF cameras are narrow)
✗$999 — most expensive in this comparison
✗ToF limited to 5.5 m — not for outdoor or long-range tasks
✗Higher power draw than stereo cameras

The Femto Mega is the choice when centimeter-accurate depth at close range is the priority — robotic grasping, inspection, and warehouse bin picking. The $999 price is justified only for applications where stereo noise is unacceptable.

#6
Legacy Research StandardStereo + ToF

Microsoft Azure Kinect DK

Microsoft · $399 (discontinued, used ~$250)

66
Range
0.25 – 5.46 m
Accuracy
< 1.4% at 1 m
FoV (H)
75° horizontal ToF
Weight
440 g
IMU
Yes ✓
ROS 2
Azure
GPU
CPU-only
Resolution
4K
ManipulationEducationIndoor SLAM
✓4K RGB + high-res ToF combination — excellent for training data collection
✓Widely used in academic papers — large codebase of examples
✓IMU integrated for body tracking and gesture work
✗Discontinued — no new stock; supply is used-only
✗440 g — too heavy for mobile robots and drones
✗x86 Linux/Windows only — no official ARM support

The Kinect DK is viable for stationary research setups where you already own one. For new projects, buy RealSense or OAK-D instead — the Kinect is discontinued and ARM support is unreliable.

Quick Comparison

CameraPriceMax RangeGPU NeededBest ForScore
ZED 2i$44920 mYes (CUDA)Outdoor Nav93
Intel RealSense D455$2996 mNoIndoor SLAM91
Intel RealSense D435i$19910 m (typ. 3 m reliable)NoIndoor SLAM87
Luxonis OAK-D Pro W$34912 mOn-deviceManipulation85
Orbbec Femto Mega$9995.5 mNoManipulation81
Microsoft Azure Kinect DK$399 (discontinued, used ~$250)5.46 mNoManipulation66

Frequently Asked Questions

RealSense D435i vs D455 — which should I buy?

D455 if you can afford $299 — it adds a global shutter (critical for fast-moving robots), a wider baseline (better depth at longer range), and higher-quality IMU. D435i ($199) is fine for slow-moving indoor robots on a tight budget. The D435i is also lighter (72g vs 120g) — use it on drones where weight matters.

Do depth cameras work outdoors?

It depends. Passive stereo cameras (RealSense, ZED) use IR illumination that gets overwhelmed by sunlight — they degrade significantly outdoors especially in direct sunlight. ZED 2i handles outdoor best (20m range, IP66, neural depth mode). OAK-D has an optional active IR projector that helps. ToF cameras (Orbbec, Azure Kinect) are generally worse in sunlight. For demanding outdoor use, combine a depth camera with lidar.

Which depth camera works without a GPU?

RealSense D435i and D455 run entirely on CPU — they work on Raspberry Pi, any x86 PC, or Jetson without using the GPU. Orbbec cameras are also CPU-based. ZED 2i requires CUDA — it won't work on a Raspberry Pi or CPU-only machine. OAK-D has an onboard VPU (Myriad X) so depth and AI inference run on-camera.

Which depth camera has the best ROS 2 support?

Intel RealSense (D435i/D455) has the most mature ROS 2 support — the `librealsense2` driver is maintained by Intel, has 100+ ROS 2 tutorials, and is well-integrated with Nav2, MoveIt 2, and Isaac ROS. ZED is second-best. OAK-D's `depthai-ros` wrapper works but has less community documentation. Orbbec's ROS 2 support is newer.

Can I use a depth camera instead of lidar for SLAM?

Yes, with caveats. Depth cameras work well for visual SLAM (RTAB-Map, ORB-SLAM3, Isaac ROS Visual SLAM) in structured indoor environments. For large outdoor environments, long corridors, or environments with featureless walls, lidar SLAM (Cartographer, SLAM Toolbox) is more reliable. Most production mobile robots combine both: depth camera for close obstacle avoidance + lidar for SLAM.

More Robot Hardware Guides

LiDAR sensors, robot arms, Jetson vs Raspberry Pi, and Isaac Sim vs Gazebo.

Browse All Guides →