Periodic Reporting for period 1 - SoftHandler (Commercial feasibility of an integrated soft robotic system for industrial handling)
Período documentado: 2019-03-01 hasta 2020-08-31
The SoftHandler Proof of Concept grant was granted to Fondazione Istituto Italiano di Tecnologia (IIT) and qbrobotics s.r.l. and borne out of the results of the SoftHands European Research Council Advanced Grant. In SoftHands, we studied the organization and control of human hands to design robotic ones. One of the main accomplishments of the SoftHands project was the development of a novel, underactuated compliant, yet resilient, hand based on the human hand synergies. This result was achieved thanks to the thorough study of the human hand carried out during the SoftHands project, which allowed to tame the complexity of hand design.
The aim of the SoftHandler Proof of Concept project was to study the feasibility of applying the SoftHand technology to industrial environments. Pick and place tasks are very common, and robots can easily execute them when the objects are orderly presented. However, rigid robots still provide unsatisfactory solutions when grasping and manipulating objects with a partially unknown geometry, or that are randomly placed in an unstructured environment. Traditional rigid robots may, indeed, drop or damage the manipulated objects and the possible occurrence of accidental impacts may also damage the robot itself. On the other hand, the compliance of the SoftHand enables an adaptive behaviour, which is particularly useful when interacting with unknown objects or unknown environments, and it is robust to impacts. For this reason, we decided to exploit the SoftHand characteristics to develop novel compliant end-effectors. To additionally enhance these features, we developed a compliant manipulator to cope with potential inaccuracies about the object position and to reduce the exchanged forces during interactions.
These features are essential for processes, which involve unstructured environments and fragile objects with a wide variety of physical and geometrical properties. Examples of this kind of tasks are bin picking, waste sorting and handling of groceries or raw food. In bin picking objects are homogeneous but often present complex geometrical shapes and are placed with random orientations. Furthermore, the presence of bin walls and other fixtures may hinder robot motion or cause undesired collisions. In grocery handling for packaging, a very common example in the recent practice of dark storing, the goods are well organized, but they come in a wide variety of shapes, textures, weights and sizes, they are often delicate and can be easily damaged by improper manipulation. In raw food handling after harvesting, as in grocery handling, the shape, weight, size, orientation and stiffness of the objects differ extensively. Furthermore, raw food is usually placed into crates without any particular organization. Finally, waste sorting is an example where irregularity and randomness of the objects are extreme, as are the physical properties (density, weight, shape, stiffness etc.) of items such as cardboard, packaging materials, glass fragments, hard pieces of scrap metal or wood. All these cases share the possible issue of inaccuracy in the detection of the object pose. This may cause a grasp to fail, or in some cases, can generate large interaction forces that can damage the robot or the object. It is relevant to highlight how these different contexts of use can present several possibilities in terms of object placement. Sometimes, also in the same context, objects can be placed randomly or in a fixed and predefined configuration. This is the case, for example, of food handling in dark stores. In this scenario, objects can be disposed in bins randomly (as e.g. courgettes or bananas) or with specific grids (as e.g. apples or peaches). Such variability is one of the leading factors for the need of versatile handling systems. Being able to automate these processes is of particular importance since they involve several aspects that could be damaging for human beings. Indeed, they are highly repetitive tasks that may force operators to work in hazardous environments like refrigerated cells or in contact with sharp or unhealthy items.
Looking at these common logistic tasks, and how humans perform them, top-down grasping approaches are usually favored for several reasons that also include the typical layout of factory lines. During top-down grasps human beings tend to completely envelop with the fingers the object to be grasped. This movement is commonly known as caging primitive, and it is the same performed also by many conventional robotic solutions. It is also worth mentioning that human beings, during the top-down approaching phase, tend to have many interactions with the environment, e.g. with the planar surface where the object is placed, in order to realize better grasps. Inspired by these observations, we developed an integrated system, named SoftHandler, capable of overcoming some of the limitations of traditional pick-and-place industrial robots, also in presence of limited vision information. For the manipulator, we adopted a parallel delta kinematics. The delta configuration is already very common in industry, ensuring fast and precise motions. The design of the robot is based on Variable Stiffness Actuators (VSAs). VSA technology allows either a soft behaviour, when robustness and adaptivity are preferred, or a rigid behaviour, when accuracy and precision are needed. Soft end-effectors can exceed he limits of traditional rigid grasping systems. Their carefully designed mechanical transmissions for safe interaction, high resilience and intrinsic adaptivity, yield gentle but firm grasp of objects with very different shape, even when the environment is partially constrained or not completely known. For these reasons, we developed an articulated soft gripper inspired by the Pisa/IIT SoftHand technology. This device has a simple design, which is particularly suited for top-down pick-and-place tasks performed e.g. with a delta manipulator. Furthermore, the gripper preserves a low level of control complexity thanks to the presence of only one motor, and a high level of robustness.
We validated the effectiveness of the SoftHandler through two experimental tests. In the first one we benchmarked its performance, comparing the SoftHandler with its rigid counterpart. We built a second robot with the same kinematics, size, power, end-effector, but different actuation mechanism. In this way we focused on the benefits of employing a compliant actuation. Results showed that the SoftHandler achieves higher grasping success rate in presence of displacement between the object and the robot end-effector, while exchanging lower interaction forces with the environment. In the second validation phase, we tested the SoftHandler in realistic scenarios, like bin picking and grocery and raw food handling. We performed a pick and place task with real objects like eggs, food packages, cabbages, etc… Results show that the SoftHandler is able to safely move these objects despite their different shape, size and weight while preserving their integrity.
The aim of the SoftHandler Proof of Concept project was to study the feasibility of applying the SoftHand technology to industrial environments. Pick and place tasks are very common, and robots can easily execute them when the objects are orderly presented. However, rigid robots still provide unsatisfactory solutions when grasping and manipulating objects with a partially unknown geometry, or that are randomly placed in an unstructured environment. Traditional rigid robots may, indeed, drop or damage the manipulated objects and the possible occurrence of accidental impacts may also damage the robot itself. On the other hand, the compliance of the SoftHand enables an adaptive behaviour, which is particularly useful when interacting with unknown objects or unknown environments, and it is robust to impacts. For this reason, we decided to exploit the SoftHand characteristics to develop novel compliant end-effectors. To additionally enhance these features, we developed a compliant manipulator to cope with potential inaccuracies about the object position and to reduce the exchanged forces during interactions.
These features are essential for processes, which involve unstructured environments and fragile objects with a wide variety of physical and geometrical properties. Examples of this kind of tasks are bin picking, waste sorting and handling of groceries or raw food. In bin picking objects are homogeneous but often present complex geometrical shapes and are placed with random orientations. Furthermore, the presence of bin walls and other fixtures may hinder robot motion or cause undesired collisions. In grocery handling for packaging, a very common example in the recent practice of dark storing, the goods are well organized, but they come in a wide variety of shapes, textures, weights and sizes, they are often delicate and can be easily damaged by improper manipulation. In raw food handling after harvesting, as in grocery handling, the shape, weight, size, orientation and stiffness of the objects differ extensively. Furthermore, raw food is usually placed into crates without any particular organization. Finally, waste sorting is an example where irregularity and randomness of the objects are extreme, as are the physical properties (density, weight, shape, stiffness etc.) of items such as cardboard, packaging materials, glass fragments, hard pieces of scrap metal or wood. All these cases share the possible issue of inaccuracy in the detection of the object pose. This may cause a grasp to fail, or in some cases, can generate large interaction forces that can damage the robot or the object. It is relevant to highlight how these different contexts of use can present several possibilities in terms of object placement. Sometimes, also in the same context, objects can be placed randomly or in a fixed and predefined configuration. This is the case, for example, of food handling in dark stores. In this scenario, objects can be disposed in bins randomly (as e.g. courgettes or bananas) or with specific grids (as e.g. apples or peaches). Such variability is one of the leading factors for the need of versatile handling systems. Being able to automate these processes is of particular importance since they involve several aspects that could be damaging for human beings. Indeed, they are highly repetitive tasks that may force operators to work in hazardous environments like refrigerated cells or in contact with sharp or unhealthy items.
Looking at these common logistic tasks, and how humans perform them, top-down grasping approaches are usually favored for several reasons that also include the typical layout of factory lines. During top-down grasps human beings tend to completely envelop with the fingers the object to be grasped. This movement is commonly known as caging primitive, and it is the same performed also by many conventional robotic solutions. It is also worth mentioning that human beings, during the top-down approaching phase, tend to have many interactions with the environment, e.g. with the planar surface where the object is placed, in order to realize better grasps. Inspired by these observations, we developed an integrated system, named SoftHandler, capable of overcoming some of the limitations of traditional pick-and-place industrial robots, also in presence of limited vision information. For the manipulator, we adopted a parallel delta kinematics. The delta configuration is already very common in industry, ensuring fast and precise motions. The design of the robot is based on Variable Stiffness Actuators (VSAs). VSA technology allows either a soft behaviour, when robustness and adaptivity are preferred, or a rigid behaviour, when accuracy and precision are needed. Soft end-effectors can exceed he limits of traditional rigid grasping systems. Their carefully designed mechanical transmissions for safe interaction, high resilience and intrinsic adaptivity, yield gentle but firm grasp of objects with very different shape, even when the environment is partially constrained or not completely known. For these reasons, we developed an articulated soft gripper inspired by the Pisa/IIT SoftHand technology. This device has a simple design, which is particularly suited for top-down pick-and-place tasks performed e.g. with a delta manipulator. Furthermore, the gripper preserves a low level of control complexity thanks to the presence of only one motor, and a high level of robustness.
We validated the effectiveness of the SoftHandler through two experimental tests. In the first one we benchmarked its performance, comparing the SoftHandler with its rigid counterpart. We built a second robot with the same kinematics, size, power, end-effector, but different actuation mechanism. In this way we focused on the benefits of employing a compliant actuation. Results showed that the SoftHandler achieves higher grasping success rate in presence of displacement between the object and the robot end-effector, while exchanging lower interaction forces with the environment. In the second validation phase, we tested the SoftHandler in realistic scenarios, like bin picking and grocery and raw food handling. We performed a pick and place task with real objects like eggs, food packages, cabbages, etc… Results show that the SoftHandler is able to safely move these objects despite their different shape, size and weight while preserving their integrity.