From 2D to 3D: DEPTRUM Expands What Robots Can Do

Today’s robot vacuums can efficiently vacuum and mop floors, avoid obstacles, map their surroundings, and navigate autonomously. These capabilities form a well-established technology platform. In real homes, however, the three-dimensional space above the floor remains largely beyond their cleaning reach.

Baseboards are a typical example. Located at the junction of walls and floors, they are often low and narrow, with varied materials, dimensions, and profiles. Conventional robot perception systems struggle to cover and precisely identify these raised boundaries, leaving them to be cleaned by hand.

Deptrum is exploring how 3D ToF visual perception can extend cleaning robots from two-dimensional floors into three-dimensional space. Depth measurements and 3D point clouds provide the data needed to perceive raised objects, understand spatial relationships, identify targets, plan paths, and execute tasks—connecting environmental perception with robotic action.

The technology uses Deptrum’s proprietary depth cameras to capture depth information and generate a 3D point cloud. Compared with depth information alone, a point cloud records the three-dimensional positions of many spatial points, giving robots a more complete basis for understanding the relationship between a target and its surroundings.

For baseboards, the robot analyzes the spatial structure of a 3D point cloud, identifies major planes such as the floor and walls, and examines their intersections to locate the target. It can then extract the baseboard’s position, direction, surface normals, and boundaries, transforming raw point-cloud data into geometrically meaningful target information.

Wall and baseboard detection demonstration

The robot can therefore determine more than whether an object is present. It can locate the target, identify the direction in which it extends, and understand its relationship with the surrounding environment. With further algorithmic processing, these 3D data support target localization, spatial reasoning, and motion planning. Combined with actuators such as robotic arms, they provide a perception foundation for automated cleaning of baseboards and other raised boundaries.

Reliable 3D Data in Complex Real-World Environments

Home environments do not present standardized targets. Baseboards vary in material, size, and distance, placing greater demands on a 3D vision system’s ability to capture useful spatial information.

Practical demonstrations show that Deptrum’s solution consistently captures valid point clouds and maintains stable target detection across different working distances and environmental conditions.

The system identifies boundaries from 3D point clouds across glossy metal, matte metal, tile, stone, PVC, and other baseboard materials. It also captures useful 3D information for baseboards measuring 4, 7, or 10 cm high and 5 mm to 2 cm thick, providing data for subsequent recognition and spatial tasks.

From Environmental Perception to Data for Robotic Tasks

Deptrum aims to move robots from sensing their surroundings to understanding space and executing tasks, using 3D visual perception to support both spatial understanding and task-oriented perception.

The value of 3D vision lies in continuously producing spatial data that can be computed, interpreted, and used to develop algorithms. Depth and point-cloud outputs from Deptrum’s cameras provide inputs for higher-level algorithms. Robot manufacturers can build target recognition, spatial localization, and path-planning capabilities around their own product designs, actuators, and task requirements, turning 3D perception into practical robotic functions.

For cleaning robots, this opens a path beyond the floor plane into three-dimensional space. Point clouds support exploration of baseboard cleaning, edge cleaning, and raised-surface cleaning, making automation possible in areas that were previously difficult to reach.

Baseboard cleaning demonstration

The same 3D perception capabilities can also support grasping, manipulation, localization, and other robotic tasks that require precise spatial information. As algorithms and actuation become more closely integrated, point clouds form a vital data link between physical environments, robot software, and real tasks.

Moving from 2D to 3D changes both the area a robot can work in and the way it perceives and understands physical space. Deptrum’s 3D ToF vision supplies high-quality spatial data for manufacturers developing new perception and task algorithms. It helps robots progress from basic environmental sensing toward spatial understanding and task-oriented perception, enabling three-dimensional cleaning and a wider range of physical tasks.