Food Handling

The food sector is claimed to have significant po- tential for the application of robots as fundamental change to productivity, product quality, and worker er- gonomics can be achieved [54。35]。 In food automation, untouched by human hand entails critical requirements for robot automation such as the need for hygienic design, operational speed, ease of programming, and cost。 In the past, these requirements had been difficult to achieve due to high throughputs, therefore requir- ing rapid grasps and fast robot motion, robust sensing for detecting object locations on conveyor belts。 High speed at high flexibility apparently is a key in the indi-

vidual handling of food objects。 Therefore, fast SCARA and parallel robots have found wide acceptance in this field。

An example of a packaging line in food produc- tion is depicted in Fig。 54。9 where cut mini  salamis are delivered in four streams per conveyor belt in ran- dom sequence。 The positions of the sausages on the translucent belt are determined by a computer vision system。 The robot picks the sausage from the mea- sured position sequentially until the gripper holds three sausages which are then placed into cavities。 With four parallel robots a maximum pick rate of 600 sausages per minute can be processed。 Key of the application is its high-speed 2-D (two-dimensional) computer vision system which feeds the robot’s path planning for col- lision-free picks at a minimum loss rate of unpicked salami [54。37]。

Recent efforts have led to customized designs of gripper systems through 3-D (three-dimensional) print- ing (additive manufacturing) which for instance in- cludes actuation through pneumatically driven bellows and low-wear metal joints。 An example of a highly ac- tuated gripper based on 3-D-printing for use in food handling is depicted in Fig。 54。10。 Additive   manufac-

Fig。54。11a,b Procedure of a bin-picking method [54。36] (a) and gripper with additional degree of freedom for reaching deep into bins (b)。 Depicted is a 2-D laser scanner on a swiveling unit for acquiring the point cloud in parallel to the robot’s motion。 The object detection itself consumes 0:5—2 s and is less time critical than the robot’s motion and grasps

turing processes seem to be perfectly suited to achieve higher flexibility in manufacturing automation [54。38]。 Numerous materials have become available for dif- ferent additive manufacturing processes so that even specific manufacturing requirements can be matched。 Initial doubts about gripper durability have been dis- pelled: lifetimes of more than 10 million load cycles for robot grippers manufactured on the basis of laser sin- tered polyamide have been reported。

Bin-Picking

Generally, industrial practice in robot workcell plan- ning aims at finding a compromise between reducing the variation of the  workpiece location  and  the cost of sensor systems to compensate for residual variation or uncertainties。 Today, nearly all parts arrive at robot workcells in a repeatable manner, either being stored in special carriers or magazines, or by being transported and oriented by vibrating devices that allow the parts to settle into a predictable orientation for proper robot grasping。 However, cost and flexibility requirements in manufacturing automation will result in reducing cus- tomized parts magazines to more universal carriers, containers, or conveyor belts。 If randomly oriented on a conveyor belt or in a carrier, parts have to be properly identified and located so that the robot can produce an collision-free grasp。

The challenge of grasping partly or randomly or- dered parts by robot has been referred to as bin- picking  and  has  been  investigated  by  numerous re-

searchers since the mid-1980s [54。39]。 Even though an abundance of approaches has been presented, only re- cently bin-picking installations have found their way into daily manufacturing in significant numbers。 Bin- picking algorithms follow a typical sequence of steps: initial point cloud data acquisition, object detection, pose estimation, collision-free path and grasp plan- ning, object grasping, and object placement。 Most methods in bin-picking assume known geometrical representations (a computer-aided design CAD-model) of the workpiece in question including the specifi- cation of admissible grasps for applying template- matching methods  [54。40, 41]。  Figure  54。11  depicts a variant of a fast template-matching method, which encompasses the following steps for detecting object poses: