These points are general. Consider the below points: 1. Access the sensor's robustness in the environment you are going to use. Some factors to look upon are its resistance to dust, moisture and temperature variations. 2. Having backup sensors are always helpful in need. Keep 2 sensors as a rule of thumb. 3. Evaluate the potential for interference between sensors. Choose sensors with minimal interference, or implement techniques to manage interference effectively, especially in scenarios where multiple robots or sensors coexist. 4. Select sensors with upgradability in mind

The dimensional footprint and weight of a sensor can differ drastically depending on the type that is chosen. Certain sensors may have different placement requirements so as to not be obstructed and to ensure reliable operation.

It is crucial to take into account the trade-offs between on-board and off-board processing while processing sensor data. On-board processing can improve response times and cut down on the requirement for communication bandwidth, but it can also use up a lot of computational power and resources. On the other hand, off-board processing may be more computationally demanding and require a steady, high-bandwidth connection to the robot.

As the development and maturation of hardware solutions progress, there are very many opportunities to implement obstacle avoidance. Some of them: Ultrasonic Sensors: Ultrasonic sensors use sound waves to detect obstacles. Infrared (IR) Sensors: Infrared sensors emit infrared light and measure the reflection or presence of infrared light from obstacles. Lidar (Light Detection and Ranging): Lidar sensors use lasers to measure distances to objects in the environment. Camera-based Vision Systems: Cameras can be used for obstacle detection by analyzing images or video frames to identify obstacles in the robot's path. Stereo Vision: Stereo vision uses two cameras placed slightly apart to mimic human binocular vision.

I think Lidar is the leading obstacle detection for industrial automation with AMRs and AGVs used for navigation on manufacturing floors. Lidar creates a 3D map of the environment, allowing the robots to accurately navigate through the space and avoid obstacles. This is probably one of the leading technologies to advance the warehouse and manufacturing environments.

Simultaneous Localization and Mapping (SLAM) refers to the problem of using various sensor inputs to map an environment and determine the location of the robot inside that environment. It is a chicken-or-egg problem because you need to know your location to generate a map and the accuracy of your location depends on your map. SLAM algorithms can use a huge range of sensors including: camera sensors, LiDARs, IMUs, depth cameras and much more. This data is often combined using an enhanced Kalman filter, which uses statistical methods to solve the chicken-or-egg problem. Humans are great at solving this problem, but most of us take that for granted.

Path planning for mobile robots involves finding feasible routes from a starting point to a goal while avoiding obstacles. This is our OBJECTIVE with path planning. It begins with representing the environment, often using grid maps highlighting obstacles and free spaces. The state space encompasses all potential robot positions, forming the basis for search algorithms. Algorithms like A*, Dijkstra's, RRT, and PRM explore this space for paths based on distance, obstacle avoidance, and energy efficiency. Obstacle avoidance requires local adjustments to navigate global planning for an overall path. Sensor data from LiDAR, cameras help robots perceive their surroundings. I think these are the most basic things about path planning.

TLDR; Traversing through graph data structure is key to path planning algorithms. For path planning, we rely on the representation of the map of the environment and choose graph based methods or sampling based methods which eventually end up using some sort of graph algorithms. The common attribute of all these algorithms is the cost function and computation time wrt increase in size of map. Here we also need to consider holonomic and non-holonomic constraint to define directions in which robot can move. Dijkstra, A*, D*, RRT are commonly used algorithms. Path planning also involves solving 2 problems i.e. global planning (static) and local planning (dynamic). In ROS1, by default navigation stack uses Dijkstra for global planning.

After determining a path, the robot must effectively track it. The planned path does not account for the full complexity of the robot dynamics and sensing, so motion control is necessary. This is typically done in two steps: low-level planning and control. Low-level planning takes the path (the high-level plan) and generates the desired velocities or accelerations to get you there. This can be done effectively with Model Predictive Control, and happens at a rate of ~10-100Hz. At the control layer, you actuate the system to track the low-level planner's commands. In practice, this is almost always a PID or cascading PID controller, but more advanced control methods exist like Control Lyapunov Functions. This is typically done at 1-10KHz.

To evaluate and benchmark one's robot's ability to navigate properly, one should use one of the established test procedures from a Standards Development Organization (SDO) such as ASTM, ANSI, or ISO. By doing this, one can leverage a test that has already been evaluated by others and report results that others can both understand, but also do themselves to validate the results. By using established test & performance standards in evaluating the robot's navigation, one puts themselves on their on path by setting their benchmark and build up from there. Also, it gives others the ability to do their own tests and confirm and validate your results. This in turn builds community trust in the robot's abilities.