The rise of SLAM specific chips behind efficient navigation and positioning of robots

Robot chips and SLAM

At present, there are various combinations of chips in robot systems, including robot specific SoC schemes, universal SoC plus module schemes, and various MCU based combination schemes. These chip combinations form the underlying hardware foundation of the robot, and various powerful functions are built on these hardware controls.
Based on hardware, robot systems also have some key points, such as intelligent algorithms, sensors, multi machine collaboration, collaborative control, etc. And with the rapid development of artificial intelligence and large models, multimodal large models based on multiple sensors are also beginning to occupy an important position in robots.
The combination of intelligent algorithms and sensors is the core of SLAM, and implementing SLAM requires two types of technologies as support: sensor signal processing (including front-end processing) and pose graph optimization (including back-end processing). These technologies are independent of sensors and depend on the algorithm processing in the back-end. From the perspective of sensors, the current SLAM technology in robots mainly applies two technical routes: visual SLAM and LiDAR SLAM. Both technical routes have many landing robot projects, and each has its own advantages.
Nowadays, 3D SLAM is being promoted, and 2D sensing can no longer meet the increasingly complex requirements of robot terminals. With 3D sensors as the core, various sensors such as LiDAR/Vision, IMU, odometer, GPS are being fused and processed, and high-performance processors are configured to achieve mapping and matching positioning functions for unmanned vehicle environments.
In order to achieve higher performance SLAM effects, upstream chip manufacturers will develop their own SLAM algorithms to hardware the algorithms at the chip level. Many sensor manufacturers also do SLAM and package it with sensor components to provide the entire navigation module to downstream device manufacturers. In this trend, there are more and more chips that will hardware SLAM.

SLAM specific chip after algorithm chip transformation

SLAM specific chips can be said to have solved many pain points in the current industry. In the past, software based 3D algorithms combined with FPGA or high-performance processors often consumed a lot of power, and the SLAM part was also time-consuming. If it involves complex scenarios such as AMR and drones, the BOM of various 3D perception accessories is high, and sensor fusion is also very complex.
After the sensor navigation algorithm is integrated into a single chip, it can effectively solve the current pain points in these industries. At present, manufacturers are already producing these products, and the related chip products are also quite representative.
Yinniu Microelectronics is about to launch the second-generation 3D SLAM system level chip NU4500, which is a 3D spatial computing system level chip that integrates 3D visual perception, AI, and SLAM hardware engine. It is currently the only single chip solution in the world that achieves the integration of the three.
This chip has an 8-core high-performance CPU and can be used as the main controller for autonomous mobile robots. It can simultaneously process 10 camera information, and can achieve multi-sensor fusion with just one chip. At the same time, the edge AI computing power of the chip configuration is further improved, up to 7.5TOPS, providing a complete edge deep learning algorithm library and solutions.
A semiconductor company specializing in robot chip design has also made remarkable achievements in the field of SLAM specific system chips, and is one of the few chip manufacturers that can simultaneously provide inertial navigation eSLAM, laser SLAM navigation, and visual navigation vSLAM chips, algorithms, and complete solutions.
The currently promoted robot SLAM SoC AM890 has a computing speed 3-8 times that of general-purpose SoC, and its overall power consumption is only one-fifth that of general-purpose SoC. This chip also hardware the SLAM algorithm, achieving independent IP research and development. It uses hardware acceleration and algorithm chip to achieve GPU and FPGA functions, reducing BOM costs.

Summary

Robot manufacturers need SLAM specific system level chips to achieve efficient mapping and positioning, and now the computing power of these chips can also cover complex navigation operations. With the expansion of the robot market and the development of more robot scenarios, the market will increasingly require SLAM specific system level chips.