RB technology stack

SLAM（ Synchronous positioning and map building ）, It refers to the moving object according to the information of the sensor , While calculating your position , While building the environment map , Solve the problem of localization and map construction when robots move in unknown environment . at present ,SLAM It is mainly used in robots 、 Unmanned aerial vehicle (uav) 、 unmanned 、AR、VR Other fields . Its uses include the positioning of the sensor itself , And subsequent path planning 、 Athletic performance 、 Scene understanding .

actually ,SLAM The algorithm itself only completes the robot positioning and map building , It is not completely equivalent to what we call navigation and positioning . The navigation here , It's actually SLAM What algorithms can't do . It's called sports planning in the industry （Motion Planning）.

Let's talk about motion planning for wheeled robots like sweepers . The basic capability required here is path planning , That is to say, it's usually finished SLAM after , There's a capability called target point navigation . Popular said , It's planning a way to get from A Point to B Point out the path , Then let the robot move past .

To achieve this process , Motion planning should realize at least two levels of modules , One is called global planning , This is a little bit like our car navigator , It needs to plan a route in advance on the map , Also have the current robot position . It's up to us SLAM The system provides . It is commonly used in the industry as A* To implement this process , It's a heuristic search algorithm , Excellent . Its most popular application is in games , Like Starcraft 、 Real time strategy games like Warcraft , They all use this algorithm to calculate the trajectory of a unit .

Of course , Just planning the path is not enough , There will be a lot of emergencies in reality , For example, a child happens to be in the way , We need to adjust the original path . Of course , Sometimes this adjustment doesn't need to recalculate the global path , The robot might just make a little detour . here , We need another level of planning module , It's called local planning . It may not know where the robot is going in the end , But it's especially good at how robots get around obstacles in front of them .

Global path planning belongs to static planning , Local path planning belongs to dynamic planning . Global path planning needs to master all environmental information , Carry out path planning according to all the information of the environment map ; Local path planning only needs real-time environmental information collected by sensors , Learn about environmental map information , Then determine the location of the map and the distribution of local obstacles , Thus, the optimal path from the current node to a sub target node can be selected .

In the global path planning algorithm , It can be roughly divided into three categories ： Traditional algorithms （Dijkstra Algorithm 、A* Algorithm etc. ）、 Intelligent algorithm （PSO Algorithm 、 Genetic algorithm (ga) 、 Strengthen learning, etc ）、 Traditional and intelligent algorithms .

Mobile robots

The general direction of robot algorithm can be divided into perception algorithm and control algorithm , Perception algorithm is generally environment perception 、 Path planning , The control algorithm is generally divided into decision algorithm 、 Motion control algorithm .

Perception algorithm

slam

Path planning algorithm

A* Algorithm ,RTT Algorithm ,Dijkstra Algorithm

Motion control algorithm

pid Control algorithm ,LQR,ADRC

Decision-making algorithm

Behavior decision algorithm or behavior control strategy is the research and development focus of the robot application field （ There are different algorithms in different application fields , Of course , It can also be completely manually controlled by people , What we often call artificial intelligence , In a narrow sense, it refers to this module ), This does not mean those simple behavioral algorithms , For example, repeat actions , The robot dances in a fixed motion , Walking without obstacles or fixed obstacle routes , These are mainly hard coded , Not involved ai, Complex behavioral decision-making algorithms mainly include fsm, Analytic hierarchy process , Decision tree , The fuzzy logic , Genetic algorithm (ga) ga, Artificial neural network ann, And specific algorithms for specific problems , Such as path planning

Mechanical arm

The motion planning of the manipulator is still different from that of the mobile robot . In the motion planning of mobile robots , Robots often use a 【 spot 】 There is , Planning for mobile robots , More is to use a variety of search algorithms , On the already built map model , Search out a path curve from the starting point to the ending point . The motion planning of the manipulator is the motion of multiple rigid bodies . When considering its motion planning , Multiple rigid bodies need to be considered at the same time （ Or joints ） The movement of , therefore , How to represent multiple rigid bodies at the same time （ Or joints ） Sports become more important .