Cobots for the Automobile Assembly Line

Intelligent assist devices (IADs) are a new class of hybrid devices for direct, physical, interaction with a human operator in shared workspaces. These devices – designed for the assembly line worker – can reduce ergonomics concerns that arise due to on-the-job physical and cognitive loading, while improving safety, quality and productivity. Cobots, a sub-set of IADs, implement software defined virtual guiding surfaces while providing some amplification of human power. They exemplify the central theme of this paper – that humans are critical in many assembly operations and ergonomics tools that enable them to perform their duties are necessary. This paper describes broad design principles for human-machine interaction in these industrial settings. Prototype industrial cobots designed for testing and validation are described. Efforts at commercializing these technologies for use in industrial settings are currently underway. 1 Why intelligent assists and cobots? The typical automotive assembly facility has been permanently transformed by the advent of robots – harsh, unsafe, conditions have been mitigated while simultaneously freeing up workers for tasks that are more enjoyable and less stressful. Well documented transformations include the body shop, where sheet metal is welded into a structure, and the paint shop, where the vehicle structure is painted. In sharp contrast, other areas have gone untouched for over three decades – pneumatic tools are still in vogue. The general assembly (GA) area, where the engine and cockpit sub-systems and seats and tires are integrated with the painted shell is an excellent example of such an area. Similarly, pneumatic material handling tools dominate in the warehousing industry. The ergonomics and productivity consequences across all U.S. industries are also well documented. U.S. industries reported that in 1995 [U.S. Bureau of Labor Statistics,1997; NIOSH/Rosenstock, 1997]: 1 Hybrid devices jointly optimize human and machine aspects to achieve superior integration of the human’s and machine’s capabilities [Karwowski, Parsaei and Wilhelm, 1988]. 1⁄2 43% of worker sustained injuries and illnesses were due to bodily reaction and exertion; 1⁄2 62% of all illness cases were due to repeated trauma disorders; and that 1⁄2 32% of cases involving days away from work resulted from overexertion or repetitive motion. The total cost to US industries of these and related problems is of the order of $13 to $20 billion annually. The impact on the manufacturing sector is also rather large. IADs generally (and cobots specifically) were created to address some of the above pressing problems. General Motors has been working with Northwestern University and the University of California, Berkeley to develop these promising solutions. More recently, Ford has embarked on parallel efforts both with Northwestern University and with Fanuc Robotics. In unrelated work, Oak Ridge National Laboratory [Deeter et al, 1997] has developed assists for munitions handling. This paper focuses on the Northwestern/General Motors cobotics work – from a broader, industrial, perspective. (For technical details on the architecture and controls please refer to other publications listed in Section 4.) It describes the underlying motivation, design principles and developments from this industral/academic collaboration. Section 2 describes design parameters while Section 3 addresses the drivers of the technology. Section 4 describes some prototypes that have been built to demonstrate the technology. The conclusion, in Section 5, describes our vision for the technology. 2 Designing assist devices for the human, the product and the process Henry Ford’s observation (top) is still very appropriate for today’s General Assembly (GA) area – which has remained, from a process perspective, essentially unchanged since his time. The tooling used tends to be mechanical in nature and is primarily powered by human and pneumatic effort. Sensing and decision making are the worker’s 2 The Fanuc device will be unveiled at the RIA/ICRA workshop on Intelligent Assist Devices at the 1999 IEEE ICRA (on 5/11/99). 2/15/99 IEEE International Conference on Robotics and Automation 1999 Page 2 of 6 responsibility. The principal reasons for not automating GA are both technical and economic. From a technological perspective, using robots for assembly in processes with high geometric dimensional variability is yet to be achieved with the reliability levels required for high volume production. Further, programming complexity grows exponentially with the number of trim options offered to the customer (e.g., leather seats, two-tone color, V6 engine, over-head console). Financially, the necessary increase in physical floor space dramatically impacts costs. In summary, the worker – with unsurpassed sensing and processing abilities – is a critical component in the assembly process. The primary concern, then, is that of the worker’s well being, given that he/she tends to tire and is susceptible to injuries resulting from cognitive and motor effort. Manual vs. Automation vs. Hybrid automation Faced with this information and the plethora of electromechanical systems available for automation, plant designers and engineers need to determine which tasks to automate. The answer, we believe, depends on the complexity of the task – as reflected by its ergonomics, the process variability and complexity and the devices available. This choice can be viewed from the perspective of efficacy, as shown in Figure 1. IADs are appropriate in situations where their capabilities (guidance and force amplification) simplify a complex task and increase the operator’s effectiveness. Automation is recommended when assembly tooling is available for high volume lines. Figure 1: Where are IADs best used? (adapted from Salvendy[1988]) We will examine this model from the dual perspectives of the human operator and the product/process characteristics and attempt to converge on ideal device characteristics that best utilize the strengths of the two. Human characteristics Most mass production lines are designed around some human attributes: physical abilities, vocational skills, and social needs. As equal opportunity laws require that jobs be designed for most anyone in the population, devices need to be designed for a performance region. While every worker’s capability to lift (static) and accelerate (dynamic) loads is different, the dynamic load (force and moment) is often the dominant concern as it leads to repetitive trauma disorders. In other cases, sideways forces (which are harder to apply than fore/aft forces) dominate. The worker’s skills (intellectual as well as hand/eye/mind coordination) directly impact the learning curve and how long it takes him to get really facile (see Figure 2). Human studies have shown that this learning curve is altered by visual, tactile, kinesthetic and auditory feedback. While simpler tasks are mastered more quickly, the learning curves for tasks of varying complexity are similar. Figure 2: Effects of feedback and learning (De Jong’s law) Humans perform motions in four stages: (1) Movement to task vicinity; (2) Primary, aiming, ballastic movement; (3) Final adjustment; and (4) Finesse (quite often, tactile). Fitts’ law [Fitts, 1954] states that movement time is longer for more difficult tasks – smaller and farther tasks take longer. While the operator’s timing for all four stages reduces with experience, inter-operator variability persists. Operators like to vary their routine. We have observed that they deliberately take different paths in order to converse with colleagues or to customize the path on every pass. In addition to giving them control on the task and the ability to compensate for process variation, providing for this variability also mitigates repetitive trauma disorders. Choices slow down operators. Hick’s Law states that movement time is proportional to the logarithm of the number of choices (MT = a + b log2n, where n = the number of choices). Devices that minimize the number of choices available to the operator positively impact productivity. Finally, we have found that getting the operator’s “buy-in” is absolutely critical to the successful use of the device. As this buy-in occurs naturally if the operator is an integral part of the design and test process, we routinely include the target operators in the design team. Product/Process characteristics Payload (i.e., payload and tooling being handled). The payload being manipulated is a prime driver in the design process. Its physical features such as mass, mass Task complexity High Low Effectiveness of tooling