Seth Allen, project manager at AdeptMobile Robots, believes that ground robotic systems often have to handle "tedious, dirty, and dangerous" tasks. In other words, robotic systems are typically used for tasks where human intervention is too costly, too dangerous, or too inefficient. In many cases, the autonomous working capability of a robotic platform is an extremely important characteristic—that is, monitoring and controlling the robot's movement from one position to the next through a navigation system. Accuracy in managing position and movement is a key factor in achieving efficient autonomous work, and MEMS (Micro-Electro-Mechanical Systems) gyroscopes provide feedback detection mechanisms, which are very useful for optimizing navigation system performance.
The Seekur robot system shown in Figure 1 is an autonomous system that uses advanced MEMS devices to improve navigation performance.
Overview of Robot Navigation
A robot's movement typically begins when a position change request is issued by the central processor, which manages the overall task progress of the robot. The navigation system initiates the execution of the position change request by developing a travel plan or trajectory. The travel plan takes into account available paths, known obstacle locations, robot capabilities, and any relevant task objectives. (For example, for a specimen delivery robot in a hospital, delivery time is critical.) The travel plan is fed into a controller, which generates motion and orientation profiles for navigation control. These profiles execute actions and processes according to the travel plan. This motion is typically monitored by several detection systems, each generating feedback signals; the feedback controller combines these signals and translates them into an updated travel plan and conditions. Figure 2 is a basic block diagram of a typical navigation system.
Figure 2. Block diagram of a general navigation system.
A key step in developing a navigation system begins with a thorough understanding of each function, paying particular attention to its objectives and limitations. Each function typically has well-defined and easily implementable elements, but it also presents challenging constraints that need to be addressed. In some cases, this can be an iterative process, where identifying and addressing limitations simultaneously reveals new opportunities for optimization. An example can clearly illustrate this process.
AdeptMobileRobotsSeekur robots
AdeptMobileRobotsSeekur2 is an autonomous robot employing an inertial navigation system (INS), as shown in Figure 3. This vehicle features a four-wheel drive system, with each wheel possessing independent steering and speed control capabilities, allowing for flexible platform movement in any horizontal direction. This capability is highly valuable for robotic vehicles in emerging applications such as warehouse delivery systems, hospital specimen/supply delivery systems, and military reinforcement systems.
Figure 3. AdeptMobileRobotsSeekur navigation system.
Positive control
The robot's body commands, or primary error signals, represent the difference between the travel plan provided by the trajectory planner and the travel progress updates provided by the feedback detection system. These signals are fed into the inverse kinematics system, which translates the robot's body commands into steering and speed profiles for each wheel. These profiles are calculated using Ackerman steering relationships, integrating tire diameter, surface contact area, pitch, and other important geometric characteristics. Utilizing the Ackerman steering principle and relationships, the aforementioned robot platform can create electronically linked steering angle profiles, similar to the mechanical rack and pinion systems used in many automotive steering systems. Because these relationships are integrated remotely, eliminating the need for mechanical axle linkages, this helps minimize friction and tire slippage, reducing tire wear and energy loss, and enabling movements that simple mechanical linkages cannot achieve.
Wheel drive and steering system
Each wheel has a drive axle, mechanically connected to a drive motor via a gearbox, and simultaneously coupled to an optical encoder—the input to the range feedback system—via another gearbox. The steering axle couples to another servo motor, which determines the wheel's steering angle. This steering axle is also coupled via a gearbox to a second optical encoder, the other input to the range feedback system.
Feedback detection and control
The navigation system uses an enhanced Kalmanfilter3 to estimate the robot's pose on the travel map by combining data from multiple sensors . The travel data on the Seekur is obtained from wheel traction and steering encoders (providing conversion) and MEMS gyroscopes (providing rotation).
Measurement range
The range feedback system uses measurements of the drive and steering axis rotations from an optical encoder to estimate the robot's position, heading, and speed. In an optical encoder, an internal light source is blocked by a disc, or the light source is illuminated onto a light sensor through thousands of tiny windows. As the disc rotates, it generates a series of electrical pulses, which are typically fed into a counter circuit. The number of counts per rotation equals the number of slots in the disc, thus the number of rotations (including decimals) can be calculated from the pulse counts in the encoder circuit. Figure 4 provides a graphical reference and relationship for converting drive axis rotation counts into linear displacement (position) changes.
Figure 4. Linear displacement relationship of the measurement range.
The encoder measurements for each wheel's drive shaft and steering shaft are combined in the forward kinematics processor using the Ackermann steering formula to generate measurement data such as heading, yaw rate, position, and linear velocity.
The advantage of this measurement system lies in its direct integration with the drive and steering control systems, allowing for precise determination of their status. However, unless a set of actual coordinates is available, the system's accuracy in determining the vehicle's actual speed and direction is limited. The main limitations (or sources of error) are tire geometry consistency (accuracy and fluctuations of D in Figure 4) and interruptions in tire-ground contact. Tire geometry depends on tread consistency, tire pressure, temperature, weight, and all conditions that may change during normal robot use. Tire slippage depends on the deflection radius, speed, and surface consistency.
Location detection
The Seekur system utilizes a variety of distance sensors. For indoor applications, the system employs a 270° laser scanner to create a mapping of its environment. The laser system measures the shape, size, and distance from the laser source of objects using energy return mode and signal return time. In mapping mode, the laser system describes the characteristics of the work area by combining scans from multiple different locations within the work area (Figure 5). This produces a mapping of object position, size, and shape, serving as a reference for runtime scanning. When used in conjunction with mapping information, the laser scanner function provides precise positional information. However, this function has limitations when used alone, including the need for downtime during scanning and the inability to handle environmental changes. In warehouse environments, personnel, forklifts, pallet trucks, and many other objects frequently change position, which can affect the speed and accuracy of reaching the correct destination.
Figure 5. Laser mapping.
For outdoor applications, Seekur uses the Global Positioning System (GPS) for location measurement (Figure 6). GPS triangulates a position on the Earth's surface using the radio signal propagation time of at least four satellites, achieving an accuracy within ±1 meter. However, these systems are difficult to operate without obstructions and can be affected by buildings, trees, bridges, tunnels, and many other types of objects. In some cases, where the location and characteristics of outdoor objects are known ("urban canyons"), radar and sonar can be used to assist in location estimation even when GPS operation is interrupted. Even so, the effectiveness is often affected by dynamic conditions, such as passing vehicles or ongoing construction.
Figure 6. GPS location detection.
MEMS angular rate detection
The MEMS gyroscope used in the Seekur system directly measures the Seekur's rotational rate about the yaw (vertical) axis, which is perpendicular to the Earth's surface in the Seekur navigation reference coordinate system. The mathematical formula used to calculate the relative heading is a simple integral of the angular rate measurements over a fixed time interval (t1 to t2).
One of the main advantages of this method is that the gyroscope connected to the robot frame can measure the actual motion of the vehicle without relying on gear ratios, gear backlash, tire geometry, or surface contact integrity. However, heading estimation relies on sensor accuracy, which depends on key parameters such as bias error, noise, stability, and sensitivity. A fixed bias error is converted into heading drift rate, as shown in the following relationship including the bias error ωBE:
Bias error can be categorized into two types: current error and condition-dependent error. The Seekur system estimates the current bias error when there is no movement. This requires the navigation computer to be able to recognize states where no position change command has been executed, while also facilitating data collection, bias estimation, and correction coefficient updates. The accuracy of this process depends on sensor noise and the time available to collect data and construct the error estimate. As shown in Figure 7, the Allan variance curve provides a simplified relationship between bias accuracy and averaging time, thus determining the relationship for the ADIS16265. The ADIS16265 is an iSensor™ MEMS device similar to the gyroscope currently used in the Seekur system. In this example, Seekur can reduce the average bias error to below 0.01°/second over 20 seconds and optimize the estimation results by averaging over a period of approximately 100 seconds.
Figure 7. Allan variance curve of ADIS16265.
The Allan variance 4 relationship also helps to gain insight into the optimal integration time (τ = t2 – t1). The lowest point on this curve is typically determined as the bias stability during operation. By setting the integration time τ to be equal to the integration time associated with the lowest point on the Allan variance curve of the gyroscope used, the heading estimation results can be optimized.
Condition-dependent errors, including the bias temperature coefficient, affect performance and therefore determine how often a robot needs to be stopped to update its bias correction. Using pre-calibrated sensors helps address the most common error sources, such as temperature and power variations. For example, replacing an ADIS16060 with a pre-calibrated ADIS16265 might increase size, price, and power consumption, but it can improve temperature stability by 18 times. For a 2°C temperature change, the ADIS16060 has a maximum bias of 0.22°/second, while the ADIS16265 only has 0.012°/second.
As shown in the following equation, the sensitivity error source is proportional to the actual change in driving direction:
Commercial MEMS sensors typically have rated sensitivity errors ranging from ±5% to over ±20%, thus requiring calibration to reduce these errors. For example, pre-calibrated MEMS5 gyroscopes such as the ADIS16265 and ADIS16135 have rated errors of less than ±1%, and can achieve even higher performance in controlled environments.
Application examples:
Warehouse inventory delivery
Warehouse automation systems currently use forklifts and conveyor systems to move materials to manage inventory and meet demand. Forklifts require direct human control, while conveyor systems require regular maintenance. To maximize warehouse value, many warehouses are undergoing reconfiguration, opening the door to the application of autonomous robot platforms. A group of robots can adapt to new tasks simply by changing the software and retraining the robot navigation system, without requiring extensive engineering work to modify the forklifts and conveyor systems. A key performance requirement in warehouse delivery systems is that robots must be able to maintain consistent travel patterns, safely perform maneuvers in dynamic environments with moving obstacles, and ensure personnel safety. To illustrate the value of MEMS gyroscope feedback for Seekur in such applications, AdeptMobileRobots experimentally demonstrated Seekur's ability to maintain repeating paths with and without MEMS gyroscope feedback (Figure 8 and Figure 9). It should be noted that GPS or laser scanning correction was not used in this experiment to study the impact of MEMS gyroscope feedback.
Figure 8. Seekur path accuracy without MEMS gyroscope feedback.
Figure 9. Seekur path accuracy when using MEMS gyroscope feedback.
Comparing the path trajectories in Figures 8 and 9, the difference in path accuracy is readily apparent. It should be noted that these experiments utilized early MEMS technology, supporting stability of ~0.02°/second. Current gyroscopes offer performance improvements of 2 to 4 times at the same cost, size, and power levels. As this trend continues, the ability to maintain accurate navigation on repetitive paths will continue to improve, creating opportunities for developing more markets and applications, such as hospital specimen/supply delivery.
Supply escort
Currently, the U.S. Defense Advanced Research Projects Agency (DARPA) continues to emphasize the greater use of robotics to enhance military capabilities in its proposals. Supply escort is one example of this application, where military escort teams are exposed to enemy threats and must move slowly along predictable patterns. Precise navigation allows robots (such as Seekur) to take on more responsibility in supply escort, reducing safety threats to personnel en route. A key performance indicator is the ability to manage GPS outages, where MEMS gyroscope heading feedback is particularly useful. The latest Seekur navigation technology was developed specifically for this environment, using MEMS inertial measurement units (IMUs) to improve accuracy and is designed to incorporate new technologies in terrain management and other functional areas in the future.
To test the positioning performance of the system with and without an IMU, outdoor path errors were recorded and analyzed. Figure 10 compares the error relative to the true path (derived from GPS) using only the range finding method with the error when combining the range finding method and an IMU within a Kalman filter. The latter's position accuracy is nearly 15 times that of the former.
Figure 10. Seekur position error using the range method/IMU (green) and using only the range method (blue).
in conclusion
Robotics platform developers have found that MEMS gyroscope technology offers a cost-effective way to improve orientation estimation and overall accuracy in navigation systems. Pre-calibrated, system-ready devices enable simple functional integration, facilitating a smooth start to development and allowing engineers to focus on system optimization. As MEMS technology continues to improve gyroscope noise, stability, and accuracy, levels of precision and control will continue to rise, opening up new markets for autonomous robotic platforms. Next-generation development of systems such as Seekur can transition from gyroscopes to fully integrated MEMSIMU/6DoF sensors. While yaw-oriented approaches are useful, the world is not flat; many other current and future applications can leverage MEMSIMU for terrain management and further accuracy improvements, with full alignment feedback and correction achieved through three gyroscopes.
