The traditional robot teaching approach, termed as “online” teaching, involves the use of teach pendants to manually control the robot through a series of waypoints to achieve the desired motion. The accuracy of the robot motion is determined via visual feedback from the operator, the skills of the operator in the use of teach pendants to move the robot, and the repeatability of the robot. This process is time consuming and the quality of the robot path is dependent on the skills of the operator. In addition, due to the need to use the robot during the teaching/programming process, it will also lead to significant downtime of the robotic work cell when a new program is required. The combination of all these factors makes the traditional robot teaching approach challenging to handle “high-mix low-volume” processes or for continuous type operations such as machining and polishing of freeform surfaces where a large number of points are required.
In order to overcome the above issues, alternative programming methods have been proposed and were summarized in (Pan et al. 2012). One useful approach to handle large work objects which require many teaching points is “off-line” programming. In off-line programming, the physical robotic work cell is replicated in a virtual environment. Subsequently, the robot path is generated based on the work object geometry and its relative placement to the robot in the virtual environment. Benefits of off-line robot programming include:

1. Reduced downtime on the production line since the physical work cell is not required during the programming of the robot
2. Increased safety factor by eliminating the need for the operator to program the robot in close proximity to the robot
3. Increased consistency of the path quality by removing the reliance on operator proficiency in online teaching

Effective implementation of off-line programming requires the virtual model of the work cell to be an accurate representation of the actual work cell. Unlike online teaching, waypoint accuracy is largely dependent on the effectiveness of the robot nominal kinematic model in determining the actual joint to end-effector pose relationship. In addition, the virtual work cell must accurately describe its real-world counterpart, in particular the relative poses of the robot, work objects, and other critical work cell elements. Failure to meet the above conditions will prevent the robot from completing the desired tasks and may even lead to collisions within the work cell. Hence, robot work cell accuracy becomes an important issue, and solutions such as robot calibration, error identification, and compensation have been proposed.
Industrial robotic manipulators generally have much better repeatability compared to its accuracy due to the emphasis of repeatability in traditional robot programming and applications (Greenway 2000), since accuracy of the waypoints is compensated via operator visual feedback in online teaching. Error sources that can compromise robot accuracy can be broadly categorized into two types, namely, (i) geometric and (ii) non-geometric errors (Mooring et al. 1991; Greenway 2000). These errors cause the actual kinematic model of the robot to deviate from the known nominal model. To achieve better accuracy, calibration of the robot is required to identify the actual kinematic model thereby compensating for the errors.
In this chapter, the focus is on methods for increasing work cell accuracy to achieve a desired tool to part motion. Firstly, in section “Literature Survey and Commercial Solutions,” a survey of existing commercial solutions to increase work cell accuracy is presented along with recent works in the academia. Following this, in section “Mathematical Preliminaries,” some mathematical notations and concepts that are required in the subsequent formulations are introduced. In section “Calibration,” a robot calibration solution is presented using the product-of-exponentials (POE) representation to identify an improved kinematic model for a robot. This method is implemented on an industrial robot and experimental results verify its effectiveness. Subsequently, in section “Compensation,” a non-model-based compensation approach is developed where improved robot accuracy in a localized region of the robot task space is achieved through the use of sensors attached to the robot end effector. In section “Calibration with Compensation Framework,” a framework to improve the robot work cell accuracy incorporating the benefits of both model-based calibration and non-model-based compensation is presented. Finally, the contents and findings of this chapter are briefly summarized in section “Summary.”