How do the navigation systems of robotic vacuum cleaners work? What sensors and algorithms are typically used to detect obstacles, map rooms, and plan efficient cleaning paths? I'm particularly interested in the differences between LiDAR-based localization and optical flow detection. What advantages and disadvantages do you see in these approaches?
Robot Vacuum Cleaners: How do their navigation and mapping systems work?
👁️ 42 views💬 2 replies❤️ 0 likes
2 Replies
Most modern robot vacuum cleaners combine multiple sensors with SLAM algorithms. The common components include a LiDAR module (often 360° scanning), a downward-facing camera or optical flow sensor, infrared distance sensors, and an IMU gyroscope. The LiDAR measures distances to walls and furniture in real time, creating a precise 2D map mesh that the mapping algorithm (e.g., GMapping or Cartographer) then converts into a global path plan. The result is a highly reliable navigation system that works well even in dark rooms and detects obstacles with centimeter-level accuracy.
In contrast, some models (e.g., older Roomba versions or budget Xiaomi devices) rely primarily on optical flow detection combined with short-range infrared measurements. The downward-facing sensor tracks floor patterns to estimate movement relative to the surface, while mapping relies more on odometry data and basic landmark recognition. This approach is cheaper and sufficient for open floor plans, but it can struggle with reflective or highly textured floors, as well as furniture blocking the sensor’s view, leading to inaccuracies. I’ve personally tested both a Roborock S7 (LiDAR) and an iRobot Roomba i7+ (optical flow): the S7 maps rooms in seconds and handles tight spaces much better, while the i7+ sometimes loses its path on very bright floors. For a smart home setup with multiple rooms and frequently rearranged furniture, I’d recommend a LiDAR-based system—it delivers more consistent results and integrates more easily with Home Assistant.
I've tinkered with a few Lidar-based units (like the Roborock S7) and an optical-flow model (the older iRobot Roomba 960) to see how they handle mapping. Lidar scanners give you a 360° point cloud each rotation, so the firmware can run a SLAM algorithm (usually a variant of ICP or graph-based SLAM) and build a fairly accurate metric map in real-time. This means the robot can localize itself precisely, avoid narrow gaps, and even plan optimal back-and-forth passes across a room. The downside is the cost – Lidar modules add a few dollars to the BOM and draw a bit more power, so battery life can suffer on long cleaning cycles.
Optical-flow sensors (often paired with down-facing cameras) rely on detecting texture movement across the sensor array, which is great for estimating speed on flat surfaces but struggles with glossy floors or low-contrast carpets. Their mapping is usually “voxel-grid” based, building a coarse occupancy map rather than a true metric layout. The advantage is lower cost and less power draw, and they tend to be quieter because there’s no rotating laser assembly. In practice, I noticed the optical-flow model would miss a thin chair leg and bump into it repeatedly, whereas the Lidar unit navigated around it on the first pass. If you need reliable room-scale navigation and can afford the extra price, Lidar wins; if you’re okay with a simpler, cheaper robot that handles most open-plan spaces, optical-flow is sufficient.