Similar to why humans have senses, a humanoid robot has sensors to be able to obtain data from its environment. This data must be processed effectively which should run in coordination with the main control system of the robot. Let’s first take a look at typical sensor types that are used in a humanoid robot.

Sensor Types in a Humanoid Robot

Force sensors: Measuring force is essential during maintaining balance, gripping, walking, reacting to external forces and adjusting its own action. For safety around humans, it is also crucial to be able to measure the force after collisions or unexpected contact, and either stopping or changing movement of joints. Just like humans, force sensors enable robots to learn to better interact with their physical environment. As we mentioned in our post about AI, one of the most studied areas in AI right now is to develop “physical” AI.

Tactile Sensors: They are used to mimic human sense of touch. It is mostly used in robotic hands, for precise gripping of objects.

Accelerometers: Also used in all types of robotic vehicles, accelerometers measure acceleration, which means increase or decrease in velocity in a given direction or multiple directions. This is crucial in determining robot’s orientation, tilting, such as vertically, sideways or in between, serving similar to an inner ear in humans. Together with gyroscopes, they are crucial in maintaining balance and stability of the robot and enabling proper walking or doing any sort of movement with its body or maintaining a certain posture.

Gyroscopes: Gyroscopes measure angular rotation. It has the similar duties as the accelerometers, and they almost always work together to obtain a clearer picture of robot’s orientation, position, turning and spinning, tilting, shocks and its state of motion in a given instant.

Inertial Measurement Unit (IMU): Accelerometers, gryoscopes and optionally magnetometers are often combined in a single unit as IMU, serving the duties mentioned above. Magnetometer serves like a compass and used in robotic vehicles more commonly.

Joint position and angle sensors: They are used to track position of joints which enables accurate measuring of the robot’s position and ensures precise movements.

Visual Sensors: A broader term than a camera, visual sensors include different types of cameras. Firstly of course a standard color camera (RGB camera) is used to see objects, faces and the environment, which is the main source of a robot’s visual data input. Other types of cameras can also be used such as depth cameras, which are used to sense depth in environments. Note that same effect can be obtained by having two regular cameras side by side, similar to how two of our eyes provide us the sense of depth. Other types of cameras that can be used are stereo, time of flight, infrared.

LIDAR (Light Detection and Ranging) sensors are used to map the environment which is very important in navigation and robot’s ability to determine the position of itself and other objects in the environment. It works by measuring the time it takes for laser pulses that sent from the sensor to return back to the sensor. Through LIDAR, which is also heavily used in autonomous vehicles, the robot can obtain detailed map of its surroundings. Lidar sensors, which can be easily fitted on autonomous cars, can be rather heavy for a humanoid robot and they can consume significant amount of power. It is also a costly component.

Ultrasonic sensors: Much simpler than LIDAR, it simply measures distance based on the reflection of ultrasound waves. It cannot be used to create highly detailed map of the envoronment like a LIDAR but it is still very useful since it can show the distance between the robot and nearby objects.

Microphones: These are for perceiving environmental sounds and speech.

Of course there can also be other types of sensors as well, as needed for a particular robot’s usage.

Sensor Data Analysis in Humanoid Robots

A humanoid robot collects data continuously from its environment while in operation. This data, when collected, is raw and may include noise.

Sensor data is multimodal, which means it comes form a lot of sensors at the same time. It is also not perfect, which includes noise. This noise needs to be filtered. All of this must also be processed in real time by combining data from multiple sensors and making meaning out of it.

Techniques and methods to process sensor data includes the ones below:

• Filtering and denoising such as Kalman Filtering
• Sensor Fusion: Combining data from multiple sensors, such as IMU and camera for a much better posture and motion control
• Aligning data from different modes
• Time synchronization
• Time series analysis, such as for motion prediction
• Sensorimotor learning: to learn from actions by reinforcement or imitations
• Anomaly detection: to detect faults

By: A. Tuter

——————————————————

Terms of use:

Copying or republishing of any type of content from this website is not allowed without written permission from us. We make dated records of our posts, keep originals of our images.

The content in this website may contain errors, inaccuracies or be incomplete. User assumes all liability as a result of usage of content in our site. Also see our Terms page for more details.