Which sensing modalities are commonly used for 3D perception in robotics, and what are typical trade-offs?

• A. Monocular cameras only; trade-offs include no depth perception.
• B. Ultrasonic sensors; trade-offs include limited field of view.
• C. Stereo vision, Structured light, Time-of-flight; trade-offs include resolution, lighting sensitivity, range, speed, and cost.
• D. Lidar, Radar, Thermal imaging; trade-offs include battery life and color accuracy.

Answer: (C)

Depth perception in robotics is achieved with depth-sensing methods that turn visual information into 3D structure. The most commonly used approaches are stereo vision, structured light, and time-of-flight.

Stereo vision uses two or more cameras to infer depth from disparity between parallel views. It can deliver high-resolution depth maps and leverages inexpensive, passive sensing, but it needs good texture in the scene and careful calibration. Performance can suffer in low-texture areas or with poor lighting, and processing disparity at high resolution can be computationally heavy.

Structured light projects a known pattern onto the scene and uses a camera to observe how that pattern deforms to recover depth. This yields precise depth at close to mid-range with relatively straightforward processing and good accuracy. However, it is sensitive to ambient light, struggles with highly reflective or non-textured surfaces, and is typically limited to shorter ranges.

Time-of-flight measures the time or phase delay of emitted light to determine distance to surfaces. It provides fast, dense depth data over moderate to longer ranges and works in a variety of lighting conditions. The trade-offs include lower spatial resolution compared to high-end stereo systems, potential interference from ambient light, and power consumption/cost considerations.

These trade-offs—resolution, lighting sensitivity, range, speed, and cost—capture why these modalities are favored for 3D perception. Other options mention modalities like monocular cues (which don’t give true depth on their own) or sensors with limited spatial detail, and while they can be useful in specific setups, they don’t represent the common, general-purpose 3D sensing triad described above.