Programming a robot to avoid falling off a cliff is a challenging task that requires a combination of sensors, algorithms, and logic. In this article, you will learn some basic concepts and techniques that can help you design a robust and safe robot navigation system.
Cliff detection is the ability of a robot to recognize and avoid steep drops or edges in its environment. This is important for several reasons. First, falling off a cliff can damage or destroy the robot and its components. Second, falling off a cliff can cause the robot to lose its orientation and location, making it harder to resume its mission. Third, falling off a cliff can pose a risk to humans or animals that may be nearby.
Most robots are not designed to absorb shocks or to survive drop tests. Falling off a cliff or edge of only a foot or two might break the robot's mechanical structure, damage its components, or at least disconnect some connectors. For that, developing a solution that allows the robot to detect cliffs and behave accordingly would be easier and much more effective than developing a robot sturdy enough to absorb shocks and withstand falling and high accelerations.
There are 2 approaches a Simple apprach that might not be very helpful based robot specs and the Complex apprach. 1. Simple apprach: Use a sensor usually IR sensor facing downwards at the edge of the robot. (Most Vacuum Cleaning robots have one like roomba) if the robot is traveling at high speed and it's huge in size this might not work due to it's inertia. (If it's a humanoid won't work either) 2.Complex apprach: a. Plane segmentaion using cameras to detect if the floor isn't at the same plane level. b. Use PointCloud data (from a lidar,Stereo cameras, depth cameras) to implment a terrain detection algorithm. Using an IMU with these algorithms also is good to compensate and elevation changes the robot might face, based on the terrain.
In robots, to have information about any parameter such as speed, position, ...etc, the robot need to be equipped with the appropriate sensor which can directly/indirectly provide useful information about the parameter that needs to be measured. In the literature, we can find so many types of sensors that can be active or passive, analog or digital. For the cliff problem, a digital sensor can be used. In the robot program since the used sensor is digital so the measurement can be either 0 (no cliff) or 1 (cliff). Based on this the designed algorithm should decide to stop moving forward and back up the robot.
FOV sufficiency with human quick WCET. LiDAR, sonar, and/or depth cameras may be used, but IFF granularity, range, repeatibility and accuracy are sufficient to reveal any cliff that could topple the mobile robot. Same holds true for upward vertical traverses.
The most accurate way to detect cliff is to use TOF sensors that works much better than IR. Number of sensors around robot improve cliff detection and have them on an angle can improve as well to stop the robot in time. However, just a simple math calculation can be backed into the logic and sensor fusing to prevent the robot to drop right after the cliff sensors detect the cliff. Location of the sensors is key to catch corner cases (especially in reverse mode) or specific cliff approaching angles. LiDAR and VSLAM mapping technologies can help to locate the cliff and react (slow speed) in the final approaching phase (false positive may happen tho)
To detect cliffs, you need sensors that can measure the distance and angle of the ground below the robot. There are different types of sensors that can do this, such as infrared, ultrasonic, laser, or camera. Each sensor has its own advantages and disadvantages, depending on the cost, accuracy, range, and reliability. You may need to use multiple sensors or fuse their data to get a more comprehensive and reliable reading.
Depending on the application and budget available, many technologies can be applied for cliff detection. It can be as simple as an infrared reflective sensor capable of detecting the height below the robot to identify edges. More sophisticated approaches can make use of RGB cameras, depth cameras, LIDAR, sonars, thermal cameras, etc. They can be used alone or combined in a sensor fusion strategy. The selection shall consider many factors including the size of the robot, the complexity of the environment, and the type of terrain. For instance, a vacuum cleaner robot could rely on infrared sensors installed on its edges. Conversely, an autonomous mining truck would need a set of different sensors used in combination to identify larger cliffs.
To program a robot to avoid falling off a cliff, you need sensors like ultrasonic, LIDAR, infrared, or cameras to detect changes in terrain elevation. Ultrasonic sensors measure ground distance, while LIDAR scans for drops in elevation. Infrared sensors detect changes in surface reflectivity, and cameras provide visual feedback. Integrating these sensors with algorithms allows the robot to detect cliff edges and adjust its path accordingly, ensuring safe navigation.
There is a wide range of sensors that can be used for cliff detection. It can be a simple infrared or ultrasonic sensor, as it can also be a depth camera or a LiDAR. Depending on the tasks of the robot and its environment, a decision can be made for the best sensor. If the robot is autonomous and able to detect obstacles in its way through a Lidar, that same Lidar could be the best sensor for cliff detection. If the robot already has integrated cameras and a control unit that runs computer vision algorithms, it would be easier to use the same cameras for the purpose of cliff detection and falling avoidance.
Cliff detection as easy it may sound its not, and not that hard also. Cliff as we see on maps are marked with sudden drop in contour. We have to find a way to find this sudden drop. One of the easiest way is to scout ahead - If you have a rover kind of bot then you can have an extension which will scout ahead of you and find if there is ground or not. It can be of a length which will give enough time for a robot to stop once cliff is detected. Another method is careful step and check. You can consider this as how we tend to walk on stairs in pitch black, we extend our foot and try and find a step then put our body pressure on it. For scouting ahead we can use Lasers, IRs, Cameras, Sonar and anything of proximity detection sensors.
Equipping the robot with suitable sensors is essential. Common sensors for cliff detection include infrared sensors, ultrasonic sensors, or specialized cliff sensors. These sensors detect changes in surface distance and provide real-time data about the environment.
Once you have the sensor data, you need to process it to determine if there is a cliff or not. This can be done using different algorithms, such as thresholding, edge detection, or machine learning. Thresholding is a simple method that compares the sensor value to a predefined limit and decides if it is above or below it. Edge detection is a more advanced method that identifies the boundaries or changes in the sensor data and calculates the slope or curvature of the ground. Machine learning is a sophisticated method that uses a trained model to classify the sensor data as cliff or not.
All you need is a camera with a CV model that can detect a lemming. If you notice lots of lemmings walking towards a specific point and then disappearing, that is likely a cliff.
An interesting thought is that there is a trade-off between the complexity of data and the type of data itself. If you use sensors such as LIDAR or camera, the processing might include slightly complex algorithms such as edge detection and so on. While if the robot contains a simple ultrasonic sensor pointing downwards at the head of the robot, a simple spike in value can be used to trigger the robot to react to the edge.
Remember, the most sophisticated solution is not required for evey case. Defining your requirement often gives you enough clarity on the required algorithm. In case of use like vaccuming robot, a normal thresholding should be sufficient. In a task like surveying, you would more likely use edge detection to estimate the ground profile. In seldom cases of such applications we use ML based approaches to infer more insights based on other onboard sensors too.
ML and NN can be very useful to detect the cliff and decide what action to take, but I recommend using an interrupt to stop the robot when the cliff is detected, after that validate the existence using computer vision or other algorithms, complex algorithms can It will take a few seconds to detect the cliff and this will be very dangerous for the robot, but if you use an interrupt link to a proximity sensor it can stop the robot at that moment.
The collected sensor data needs to be processed to make informed decisions. Implement algorithms that analyze the sensor inputs to identify potential cliffs or drop-offs. Signal processing and data filtering techniques can enhance the accuracy of cliff detection.
After you detect a cliff, you need to decide how to react to it. This depends on your robot's goals, capabilities, and constraints. You may have different options, such as stopping, turning, backing up, or jumping. You need to program your robot with a logic that can choose the best option based on the situation and the sensor data. You also need to consider the trade-offs between safety, speed, and efficiency.
The reaction to the cliff can vary depending upon the distance at which a cliff can be detected as well as the speed the robot is traveling. A cliff detected with significant advanced notice can simply plot a new trajectory to continue it's mission. A cliff detected with a short distance will require the robot to stop or implement some other tactic to prevent traveling over the cliff. There could be a backup solution, for example a high friction feature on the bottom of the robot to arrest it's forward speed or an actuated mass to shift the robots center of gravity to avoid a fall.
Upon detecting a cliff, the robot must execute appropriate actions to avoid falling. This may involve immediate braking, altering its path, or triggering emergency maneuvers. The reaction strategy depends on the specific robot design, its mobility capabilities, and the nature of the environment.
Now comes CLIFF AVOIDANCE algorithms. Basically obstance avoidance algorithm with some tweaks. So many ready made stuff available. Look into Bug, Tangent Bug algorithms.
Once a cliff is detected, the robot must determine its response based on its objectives, abilities, and limitations. Possible reactions include stopping, changing direction, reversing, or adjusting its speed. The robot's programming should incorporate logic to assess the situation and select the most appropriate action considering factors like safety, speed, and efficiency. It's essential to balance these considerations to ensure the robot navigates effectively while avoiding potential hazards.
Pode ser mencionado que atualmente uma das melhores maneiras de se reagir aos dados quando decidido que se trata de um penhasco, seria a tomada de decis?o a partir de um aprendizado de máquina/IA.
Finally, you need to test and improve your system to make sure it works as expected and can handle different scenarios. You can do this by simulating your system using software tools or by testing it on real terrain with different cliffs. You should monitor and evaluate your system's performance, accuracy, and reliability. You should also identify and fix any errors or bugs that may occur. You should also try to optimize your system by reducing the complexity, cost, and power consumption.
Testing is critical to validate the effectiveness of the cliff detection system. Conduct controlled experiments in various environments, simulating cliff-like conditions. Continuous improvement involves refining algorithms, adjusting sensor configurations, and addressing any false positives or negatives encountered during testing.
Testing and improving the system involves unit testing components like sensors and algorithms for functionality and accuracy. Integration testing assesses interactions between components, while simulation and field testing evaluate performance under different scenarios. Iterative improvements are made based on testing results, including refining algorithms and adjusting parameters. Benchmarking against industry standards gauges performance, and user feedback provides insights for practicality and usability. This systematic approach ensures continuous enhancement of the system's reliability, accuracy, and overall performance within operational constraints.
Robotic systems are mostly specific to an environment. So, it will be best to test them in its intended operating environment.
If you're rich enough, make a few hundred of your robots and roll them of a cliff one by one. Else, isolate the detection sensor and simulate a cliff in a controlled environment. If you have time, include all possible variables and do a Matlab simulation maybe...
El mejor escenario para realizar pruebas es uno simulado, actualmente existen motores gráficos con unreal engine 5 en los que se pueden recrear las físicas del terreno de una manera bastante óptima
Cliff detection is an invaluable skill for robots that operate in outdoor or indoor environments with uneven or dynamic surfaces. Vacuum cleaning robots can use it to avoid falling off stairs or ledges, exploration robots can traverse rough or unknown terrain without getting stuck or lost, rescue robots can navigate through disaster zones or hazardous areas without harming themselves or others, and entertainment robots can perform stunts or tricks without crashing or breaking. In short, cliff detection can help robots of all kinds safely and efficiently navigate their environment.
if (robot_is_falling_off_a_cliff) { don't do that } if (does_it_anyway) { //add code here to handle the situation… //call a robo medevac? }
The Roomba, as the most common household robot, is an obvious example. The cliff sensing technology uses simple infrared sensors to detect stairs.
Examples of cliff detection applications include robotic vacuum cleaners avoiding stairs, drones navigating varied terrains, and autonomous vehicles preventing accidental drops from elevated surfaces. The technology is versatile and applicable across a range of robotic systems.
Autonomous Vehicles: Self-driving cars and drones utilize cliff detection to avoid driving off cliffs, bridges, or other elevated surfaces. Search and Rescue Robots: Robots deployed in search and rescue missions, such as after natural disasters or in hazardous environments, need to navigate through rubble. Agricultural Robots: Agricultural robots used for tasks such as crop monitoring. cliff detection plays a vital role in ensuring the safety and efficiency of robots operating in various environments, from urban settings to remote wilderness areas by detecting and avoiding potential hazards, these robots can carry out their tasks effectively while minimizing the risk of accidents or damage.
Es gibt einige Anwendungsbereiche wie z.B. Autonome Mobile Roboter in der Logistik, Intralogistik, Agrar und Landwirtschaftliche Nutzung oder zum Materialtransport
Also consider the dynamics and statics of the system. Between the placement of sensors, sensing distance, wheels, you have CG , coefficient of fiction of tires, timing of the electro- software system.
Robotics Engineer | MSc in Mechanical Engineering at National University of Singapore
For a differential drive robot, that is navigating in an environment with corners and edges, such as a big table, a recommendation would be to use 2 drive wheels and two caster wheels, one in front and one at back. Without the use of sensors to detect cliffs, an IMU sensor can be used to detect the orientation of the robot in case it dangles of the edge slightly. The drive wheels being in the middle would assist in getting the robot back on the platform. This design would do away with the need of elaborate sensing at the expense of proper mechanical calculation of drive motors, mass properties and chassis design.
I'd like to highlight a practical application of cliff detection in consumer electronics, particularly in home cleaning robots. Cliff detection is crucial in home cleaning robots, ensuring they don't fall down stairs. Employing IR sensors, these robots detect when no ground is present, triggering them to back away from potential drops. Adding ultrasonic sensors improves detection accuracy and adds a layer of redundancy. This combination safeguards the robots in multi-story homes, showcasing the vital role of sensor technology in preventing accidents and enhancing robot safety.
One of the major things to consider in detecting a cliff is the placement of the sensors. Based on the shape of the robot, the sensors need to be placed such that the cliff is detectable reliably. Distance and angle sensors need to obtain readings to determine whether a cliff exists before the robot moves over that position. This can be accomplished by placing a camera at a certain height so the ground is visible or an IR sensor that extends past the front of the robot.
First, consider any robot building system consists of three main subsystems. - Mechanical System (contains mechanisms, chassis, frame). - Electrical System (consists of power distribution system and other necessary electrical connections). - Control system (consists mainly of inputs+processor/controller+outputs). To avoid the cliff you must design your robot mechanical system to suit rough environments, use a precise suspension mechanical system and encoder feedback on your drive to create a locomotion system. - Detecting cliff in high processing, use image processing techniques with a camera. - Low processing applications, use Ultrasonic sensors / IR proximity sensors or both and combine results with ucontroller to take proper actions.