# Visual SLAM vs LiDAR SLAM for indoor robots > How an indoor robot maps a building and finds itself in it, where cameras and laser scanners each fail, and how our delivery robot does it. Page: https://www.whitehatdesignstudio.com/guides/visual-slam-vs-lidar-slam/ Published by: WhiteHat Design Studio Last updated: 7 November 2026 The short answer SLAM lets a robot build a map and work out where it is on that map at the same time. Visual SLAM does it with cameras, which cost less and see detail but need light and texture. LiDAR SLAM does it with a laser scanner, which measures distance directly but costs more and can miss glass. By the engineers at WhiteHat Design Studio. Published 7 November 2026. ![Auto Vendi driving along an indoor corridor while a person walks ahead of it.](https://www.whitehatdesignstudio.com/_astro/autovendi-corridor.BsRN_UHd_Z1R8UI8.jpg) Auto Vendi in a corridor navigation test. ## What SLAM does SLAM stands for simultaneous localisation and mapping. A survey in IEEE Transactions on Robotics describes it as building a model of the surroundings, the map, while estimating where the robot is within it. The robot needs both at once: it cannot place itself without a map, and it cannot draw the map without knowing where it stands. Visual odometry is one part of the job. It estimates how the robot moved from one camera frame to the next. Small errors add up, so on its own it drifts. SLAM corrects the drift by recognising places it has already seen. ## Cameras or a laser scanner | What matters | Visual SLAM | LiDAR SLAM | | --- | --- | --- | | Sensor | Cameras: one, a stereo pair or a depth camera | A laser scanner | | Sensor cost | Lower | Higher | | Needs | Light, and surfaces with visible texture | Surfaces that send the laser back | | Weak spot | Plain walls and low light | Glass, mirrors and polished metal | | Also gives you | Images the robot can use for object detection | Direct distances to everything around it | A 2024 study in Applied Sciences compared ORB-SLAM3, a visual system, with SC-LeGO-LOAM, a LiDAR one, on mobile robots. Both placed the robot well, and both suffered where there was little texture. The visual system used more processor time. The authors judged LiDAR better suited to the outdoors and visual SLAM strong indoors. ORB-SLAM3 itself shows how much a camera can do: it works with one camera, a stereo pair or a depth camera, with or without an inertial sensor, and it keeps several maps and joins them when it returns to a place it knows. ## Why glass matters Laser light mostly passes through glass, or reflects away from the sensor, so a LiDAR often sees no wall there at all. Foster and colleagues showed in 2013 that standard occupancy grid maps tend to delete glass obstacles, leaving what looks like open floor where a glass wall stands. Later work, such as Cartographer_glass, detects glass and adds it back to the map. If your building has glass doors, glass walls or a glazed lobby, plan for it before you choose the sensor. ## What a building does to a robot - **Long corridors look alike.** One stretch of corridor can look much like the next, which makes it hard to recognise a place. - **People move through.** The map has to keep the walls and ignore the people. - **The light changes.** Morning sun, evening lamps and a dark store room are three different worlds to a camera. - **Doors open and close.** A route that was clear an hour ago may not be clear now. ## How our robots do it Auto Vendi uses visual SLAM. It builds the map as it moves, plans its route and avoids obstacles as they appear, and YOLOv8 object detection runs alongside the mapping. Gazebo simulation is part of how it was built. Its published tests are indoors, in corridors and a lobby. HARP drives itself on a predefined map, using visual odometry and SLAM, and a gamepad takes over when an operator needs it. ## How to choose - **Glass, mirrors or dark spaces:** LiDAR, or LiDAR with cameras. - **A tight budget in a well-lit, textured building:** cameras. - **The robot has to recognise objects anyway:** cameras are already on board, and they can do both jobs. Then take the robot to the real building on its worst day: low light, people walking past, every door open. Simulation finds the faults in your logic. Only the building finds the faults in your assumptions. The [Auto Vendi case study](https://www.whitehatdesignstudio.com/work/auto-vendi/) shows the whole system, from the phone app to the locked compartment, and [our hardware practice](https://www.whitehatdesignstudio.com/services/hardware/) builds robots like it. ## Where the answer comes from - [HARP](https://www.whitehatdesignstudio.com/work/harp/) Navigates a mapped building, tracks faces and holds a conversation with an onboard language model, built around a Raspberry Pi 5. - [Auto Vendi](https://www.whitehatdesignstudio.com/work/auto-vendi/) Maps its route, avoids obstacles and delivers in a locked compartment that the phone app releases. Practices - [Hardware](https://www.whitehatdesignstudio.com/services/hardware/) Machines that survive the real world. - [AI](https://www.whitehatdesignstudio.com/services/ai/) Intelligence that runs where the work is. A hardware, C AI The depth camera is the sensor visible under the lid in the photograph. ## On film Navigating an indoor corridor. A clip from [the log](https://www.whitehatdesignstudio.com/log/), without sound. [Auto Vendi, corridor navigation. Video without sound, 15 seconds.](https://www.whitehatdesignstudio.com/media/clips/autovendi-corridor.mp4) 0:15, no soundAuto Vendi drives down a long indoor corridor. A person holding a phone walks beside it, then past it. ## Questions people ask next ### What is the difference between visual odometry and SLAM? Visual odometry estimates how the robot moved between one camera frame and the next. Small errors add up, so it drifts. SLAM keeps a map as well, and corrects the drift when it recognises a place it has seen before. ### Can a robot use cameras and LiDAR together? Yes. Combining sensors lets each one cover the other's weak spot, such as glass for LiDAR and plain walls for cameras. The cost is more hardware, more computing and more calibration. ### Do I need SLAM if the building never changes? Usually yes. Even a fixed building has people, doors, carts and changing light, and the robot still has to know where it is on the map. ## Sources - [Cadena and others, Past, present, and future of simultaneous localization and mapping, IEEE Transactions on Robotics, 2016](https://arxiv.org/abs/1606.05830) - [Campos and others, ORB-SLAM3](https://arxiv.org/abs/2007.11898) - [Zhao, Hong and Huang, Comprehensive performance evaluation between visual SLAM and LiDAR SLAM for mobile robots, Applied Sciences, 2024](https://doi.org/10.3390/app14093945) - [Foster, Sun, Park and Kuipers, VisAGGE, visible angle grid for glass environments, ICRA 2013](https://web.eecs.umich.edu/~kuipers/papers/Foster-icra-13.pdf) - [Weerakoon and others, Cartographer_glass, 2D graph SLAM using LiDAR for glass environments](https://arxiv.org/abs/2212.08633) - [Ultralytics, YOLOv8](https://docs.ultralytics.com/models/yolov8/) ## More guides - [How does drone detection work?](https://www.whitehatdesignstudio.com/guides/how-drone-detection-works/) - [STM32 vs ESP32: which should your product use?](https://www.whitehatdesignstudio.com/guides/stm32-vs-esp32/) - [Edge AI or cloud AI: where should your model run?](https://www.whitehatdesignstudio.com/guides/edge-ai-vs-cloud-ai/) [All the guides](https://www.whitehatdesignstudio.com/guides/) ## Tell us what has to work. One sentence is enough. An engineer will reply. [Talk to an engineer](https://www.whitehatdesignstudio.com/contact/) [WhatsApp an engineer](https://wa.me/923112721007?text=Hello%20WhiteHat%20Design%20Studio%2C%20I%20would%20like%20to%20talk%20to%20an%20engineer%20about%20my%20project.)