Force control and flexible touch become core needs. What kind of embodied sensing system should robots create?

Enthusiasts.com reported (Text/Mo Tingting) Embodiment refers to the ability of intelligent systems or machines to interact with the environment in real time through perception and interaction, and is also a branch of development. 2024 is considered the first year of the development of the embodied intelligent industry, and related industries are gradually developing, for example.

Industry insiders believe that embodied intelligent robots are the ultimate form of artificial intelligence. Embodied intelligent robots refer to robots that can interact with the environment, autonomously plan, make decisions, act, and perform tasks like humans. Before the upgrade of embodied intelligence is truly completed, it will play a key role in this process.

So, what kind of demand does embodied intelligence have for sensors? Let's start with the perception function.

Perception includes vision, hearing, taste, touch, etc. Embodied intelligent robots require different sensors for different perceptions, such as CMOS for visual perception, Mic vibration sensors for hearing, force sensors for pain, etc., IMU for vestibular perception, ultrasonic distance sensors for proprioception, and sensors that can realize corresponding functions for taste, touch, and smell, such as tactile arrays.

Only a complete set of embodied sensing systems can enable robots to achieve true embodied intelligence, and only through the transmission of perception can the guidance of motion control be completed.

Liu Yuanyuan, head of the bionic robot BU of Tiantai Robot, believes that sensing will cause a paradigm change in motion control, which includes changes from vision, motion path planning, alignment of the five senses, and prediction of the next frame of motion control. For example, when grasping an object, the sensor will first be used to determine the direction, then the size of the object, adjust the reach, etc., and then adjust the grasping force through the force sensor to complete a series of actions.

At present, force control and flexible tactile control are the core requirements in the perception of robots. Previously, in a video, the Tesla humanoid robot Opmus Gen2 was shown grasping and delivering eggs. To achieve a steady hold without crushing the eggs, it is necessary to rely on a tactile system, especially a flexible tactile sensor. Flexible tactile

sensors are divided into: piezoresistive (), piezoelectric, type, type, etc., and capacitive types occupy about 50% of the market share. The most critical thing for current flexible touch control is to solve the problems of pressure and sliding, that is, to solve the problem of not crushing it and holding it firmly, followed by the requirements of shape, material, temperature, etc.

In 2023, Pasini released the humanoid robot Tora, which is equipped with more than 100 multi-dimensional tactile sensors on both hands. The official website shows that the multi-dimensional tactile sensor PX-6AX has a force control accuracy of up to 0.01N, which means that the minimum measurable force is 0.01N. The sampling frequency is 1000HZ, which can achieve multi-axis output and multi-dimensional deployment. It can measure 15 physical quantities such as pressure, force, and sliding, and can capture sub-millimeter-level tiny deformations on the contact surface.

Figure: PX-6AX functional analysis (Pacinian)

In terms of force control, force sensors are key. According to the measurement dimension, force sensors can be divided into one to six-dimensional force sensors, which can simultaneously measure three forces (FX, FY, FZ) and three moments (MX, MY, MZ) in the neutral coordinate system (OXYZ). At present, the robot force control solution is more applied to the six-dimensional force sensor with the highest dimension. According to the different sensing elements, the six-dimensional force sensor can be divided into strain gauge type, type and piezoelectric/capacitive type. Yuli Instruments

displayed the M37XX series of six-axis force sensors that can be used for robots and general testing on its official website. According to the introduction, this is a matrix decoupled six-axis force sensor with a diameter of 15 to 135mm, an accuracy of 5 to 6400N, and 0.5 to 320Nm. In other aspects, the M37XX series has nonlinearity and hysteresis less than 0.5%FS, and crosstalk less than 2%FS.

In the application of robots, in addition to accuracy, weight and price are one of the key factors to consider force sensors. In the future, as robots need to master more skills, not only force control and flexible tactile control, other sensors such as vision will need to be iterated to meet the needs of different application fields.