1. Sensor failure
In principle, no sensor is perfect. For example, if an AGV (Automated Guided Vehicle) is in front of a completely transparent glass panel, infrared, lidar, or vision-based solutions might fail to detect obstacles because light can pass directly through the glass. This is where ultrasonic sensors come in to detect obstacles. Therefore, in practical applications, we definitely need to combine multiple sensors, cross-validate the data collected by different sensors, and fuse the information to ensure the AGV operates stably and reliably.
In addition to these, there are other modes that may cause sensor failure, such as ultrasonic ranging, which generally requires an ultrasonic array. If the sensors in the array work simultaneously, they are prone to mutual interference. The light wave emitted by sensor A is reflected back and received by sensor B, resulting in incorrect measurement results. However, if they work one by one in sequence, the sampling period of the ultrasonic sensor is relatively long, which will slow down the overall acquisition speed and affect real-time obstacle avoidance. This requires that both the hardware structure and the algorithm be well designed to maximize the sampling speed and reduce crosstalk between sensors.
For example, if an AGV needs to move, it generally needs a motor and a driver. During operation, these components can cause capacitor compatibility issues, which may lead to errors in sensor data acquisition, especially with analog sensors. Therefore, during implementation, it is necessary to keep the motor driver and other equipment, the sensor acquisition section, and the power and communication section isolated to ensure that the entire system can work properly.
2. Algorithm Design
Many of the algorithms mentioned earlier did not fully consider the kinematic and dynamic models of the AGV itself during the design process. The trajectories planned by such algorithms may be kinematically impossible to achieve, or they may be kinematically achievable but extremely difficult to control. For example, if an AGV's chassis is based on a car structure, it cannot turn on the spot at will. Even if the AGV can turn on the spot, its motors may not be able to execute a large maneuver. Therefore, during the design phase, it is crucial to optimize the AGV's structure and control, and the feasibility of obstacle avoidance schemes must be considered.
Then, when designing the entire algorithm architecture, we need to consider that in order to avoid or prevent injury to people or the AGV itself, obstacle avoidance is a high priority task, even the highest priority task, and the priority of its own operation and control of the AGV should also be given. At the same time, the algorithm must be fast enough to meet our real-time requirements.
In summary, in my opinion, obstacle avoidance can be seen as a special case of autonomous navigation planning for AGVs. Compared with overall global navigation, it has higher requirements for real-time performance and reliability. Furthermore, locality and dynamism are its characteristics, which are things we must pay attention to when designing the entire hardware and software architecture of AGVs.
