This paper describes the development of a novel vision-based modeling and grasping system for three-dimensional (3D) objects whose shape and location are unknown a priori. Our approach integrates online computer vision-based 3D object modeling with online 3D grasp planning and execution. A single wrist-mounted video camera is moved around the stationary object to obtain images from multiple viewpoints. Object silhouettes are extracted from these images and used to form a 3D solid model of the object. To refine the model, the object's top surface is modeled by scanning with a wrist-mounted line laser while recording images. The laser line in each image is used to form a 3D surface model that is combined with the silhouette result. The grasp planning algorithm is designed for the parallel-jaw grippers that are commonly used in industry. The algorithm analyses the solid model, generates a robust force closure grasp, and outputs the required gripper position and orientation for grasping the object. The robot then automatically picks up the object. Experiments are performed with two real-world 3D objects, a metal bracket and a hex nut. The shape, position and orientation of the objects are not known by the system a priori. The time required to compute an object model and plan a grasp was less than 4 s for each object. The experimental results demonstrate that the automated grasping system can obtain suitable models and generate successful grasps, even when the objects are not lying parallel to the supporting table.

Pneumatic actuators are low-cost, safe, clean, and exhibit a high power to weight ratio. In this paper a novel servo pneumatic system based on a miniature cylinder with a 9.5 mm bore size is presented. Four low-cost 2-way proportional valves are incorporated to provide greater design flexibility than the traditional single 4-way servo valve solution. A nonlinear system model is developed and validated using open-loop experiments. The use of bipolynomial functions to model the valve flow rates provides a more accurate solution than the commonly used nozzle flow equations. A novel multiple-input single-output nonlinear position control law is designed using the backstepping method. The stability analysis includes the effects of friction modeling error and valve modeling error. In experiments with a 1.5 kg moving mass, the new control law produced maximum tracking errors of plusmn0.5 mm for a 1 Hz sine wave trajectory, and steady-state errors within plusmn0.05 mm. The tracking errors are 82% less than those produced by a linear controller

A novel solution to the three-dimensional dynamic human-robot collision problem is presented. Sphere-based geometric models are used for the human and robot due to the efficiency of the distance computation. The collision avoidance algorithm searches for collision-free paths by moving the end-effector along a set of pre-defined search directions. An optimization method is employed to select the search direction that balances between the robot approaching its goal location, and maximizing the distances between the human and robot models. The optimization incorporates predictions of the motions of the robot and human to reduce the negative effects of a non-instantaneous robot time response. The robot prediction is based on a transfer function model of its experimental time response at the joint level. The human prediction is performed at the sphere level using the weighted mean of past velocities. Predicting at the sphere level eliminates the difficulty introduced by the limbs moving in different directions. After describing the collision avoidance algorithm, a human walking towards a moving Puma robot arm is simulated. Captured motion data is used to make the human motion realistic. Monte Carlo simulations using 1000 random human walking paths passing through the robot workspace are used to evaluate the algorithm. The algorithm prevented all collisions due to the robot. The algorithm is deterministic and efficient enough to be used in real-time. On a 1.8 GHz Pentium IV PC, a 40 Hz sampling rate was achieved

The development of a system for automatically locating and tracking a human in the vicinity of a robot is described. The system consists of multiple passive infrared (PIR) sensors, two color cameras, a pair of microwave sensors and a pair of PCs for data collection, signal processing and data fusion. The cameras are treated as individual sensors rather than a stereo pair to minimize the affect of occlusion by the robot. The area around the robot is subdivided into an occupancy grid with 0.5m by 0.5m cells. A data fusion algorithm, based on Dempster-Shafer evidence theory, is used to estimate the probability of human occupancy for each cell. This information is used to estimate the human’s location. A novel concept termed a “protective cell” is introduced to further increase the human’s safety in the presence of sensor uncertainty. Experimental results are included demonstrating the system’s effectiveness.

In this paper, we introduce unilateral fixtures , a new class of fixtures for sheet-metal parts with holes. These fixtures use cylindrical jaws with conical grooves that facilitate part alignment; each jaw provides the equivalent of four point contacts. The fixtures are unilateral in the sense that their actuating mechanisms are restricted to one side/surface of the part, facilitating access to the other side/surface for assembly or inspection. We present a two-phase algorithm for computing unilateral fixtures. Phase I is a geometric algorithm that assumes the part is rigid and applies two-dimensional (2-D) and three-dimensional (3-D) kinematic analysis of form closure to identify all candidate locations for pairs of primary jaws. We prove three new grasp properties for 2-D and 3-D grips at concave vertices and define a scale-invariant quality metric based on the sensitivity of part orientation to infinitesimal relaxation of jaw position. Phase II uses a finite element method to compute part deformation and to arrange secondary contacts at part edges and interior surfaces. For a given sheet-metal part, given as a 2-D surface embedded in 3-D with e edges, n concavities and m mesh nodes, Phase I takes O(e+n/sup 4/3/log/sup 1/3/n+glogg) time to compute a list of g pairs of primary jaws ranked by quality. Phase II computes the location of r secondary contacts in O(grm/sup 3/) time. Note to Practitioners-This paper was motivated by the problem of holding sheet-metal parts for automobile bodies but it also applies to other sheet-metal components that have cut or stamped holes. Existing approaches to fixturing such parts generally have contacting mechanisms on both sides of the sheet that restrict access for welding or inspection. This paper suggests a new approach using pairs of grooved cylinders, activated from only one side of the part (hence "unilateral"). These cylinders mate with opposing corners of holes in the sheet and push apart to hold the sheet in tension, thus acting as both locators and clamps. In this paper, we mathematically characterize the mechanics and conditions for a unilateral fixture to hold a given part. We then show how such fixtures can be efficiently computed; this can allow a computer-aided design (CAD) system (with finite element capability) to automatically generate and propose unilateral fixtures for a given part. Preliminary physical experiments suggest that this approach is feasible but it has not yet been incorporated into a CAD system nor tested in production. In future research, we will address the design of unilateral fixtures that hold two or more parts simultaneously for welding.

Pneumatic servo actuators have the benefits of low-cost, cleanliness and a high power-to-weight ratio. However, their relatively poor accuracy prevents them from competing with electro-mechanical systems when higher accuracy is needed. The cause of the steady-state error for a pneumatic servo system with an open-center servo valve is investigated. Full nonlinear and linearized plant models are presented. An effective friction compensation method is introduced which can be added to any control strategy. When combined with a novel PVA/PV control approach, a steady-state accuracy of /spl plusmn/0.01mm was verified in experiments. This is a tenfold improvement over previously reported experimental results for such systems. This performance is achieved for both vertical and horizontal movements with payloads ranging from 0.3 to 11.3kg, without re-tuning the controller.

The planning of optimal grasps is an important problem in robotics which has been investigated by many researchers. The large number of available methods has made it difficult to discern those which plan a grasp with good overall performance, i.e., one with high strength, insensitivity to positioning errors, and ease of computation. In this paper, a new grasp planning method is introduced and compared to three existing planning methods using three such metrics. A new metric for measuring the sensitivity of a grasp to positioning errors is also introduced. Since grasp planning is much simpler in 2D, and 2D grasps are applicable to many 3D objects, the four methods involve only a 2D analysis. The methods are applied to a set of six polygonal objects, ranging from 3 sided to 74 sided, and their overall performance is compared. The benchmarking procedure is readily applicable to other grasp planning methods.

The high power-to-weight ratio and low cost of pneumatic actuators makes them attractive for robotics applications, however achieving fast, accurate position control with them is difficult. Furthermore, tuning the controller (and/or obtaining process models) requires extensive expertise and time. In this paper, a novel automatic tuning methodology for the accurate position control of pneumatic actuators is described. The methodology combines off-line model based analysis with online iteration. Experiments performed on three actuators with distinct open-loop dynamics verified the auto-tuner's effectiveness. The auto-tuning required 1/50 of the time needed for manual tuning and only nonexpert supervision. The performance of the auto-tuned actuators is comparable to that achieved by other researchers using more complex and expensive hardware.

A novel grasping strategy and gripper for fixturing in 3D is presented for the robotic fixtureless assembly application. The goal of the strategy is to accurately immobilize a part in the presence of initial robot and part positioning errors. The grasping strategy expands a previously developed 2D theory into 3D and is implemented on two automotive parts using a multi-degree of freedom gripper. The gripper is able to fixture a variety of parts and the only change is reconfiguration of the computer controlled axes. To fixture a sheet metal part, the fingers are placed within holes of the part and moved until the desired set of contact locations is achieved. The fingers are grooved at fixed angles such that the edge of the sheet metal part can be held within the grooves. Three fingers and six frictionless point contacts are used for each part. A computer algorithm is described that solves for suitable contact locations based on the part geometry. The algorithm was implemented and tested on two Buick sheet metal parts from the front fender assembly. Twenty five trials were performed for each grasp. The standard deviation of the part location prior to being grasped was 0.43 mm. After being grasped, this was reduced to 0.01 mm.

The development of a fast, accurate and inexpensive position controlled pneumatic actuator that may be applied to a variety of robotic applications is described. A novel PWM valve pulsing algorithm allowed on/off solenoid valves to be used in place of costly servo valves. A linear process model was obtained from experimental data using system identification. A PID controller with added friction compensation and position feedforward was successfully implemented. A worst case steady state accuracy of 0.21 mm was achieved with a rise time of 180 ms for step inputs from 0.11 mm to 64 mm. Following errors to 64 mm s-curve profiles were less than 2.0 mm. The controller was robust to a six fold increase in the system mass. The actuator's overall performance is comparable to that achieved by other researchers using servo valves.

A novel approach to grasping with an end effector for the purposes of fixtureless assembly is presented. The grasping strategy is based on having the final position of the contacts determined by specific regions of the object geometry. The fingers are placed within limited spaces of the object and moved using frictionless contacts until motion ceases. The limited spaces usually take the form of concave edges or holes in the object. This strategy allows the positioning error to be determined by the accuracy of the part and is independent of the accuracy of the robotic manipulator. A new method for finding form closure is introduced based on maximizing the distances between contacts. The grasping strategy allows deterministic positioning of the object and also provides a means of convergence to these holding points. Testing was done in the plane with three fingers for several cases to show the sensitivity of the grasps to part geometry. The results show the position error is dependent on local shape and was reduced from 1 mm to 0.1 mm for several cases.

Two control problems encountered in robotic sheet metal assembly are addressed in this paper. They are the control of vibration when handling the sheet metal parts and the control of the contact state between the parts during assembly. For the first problem, a learning extremum controller (LEC) is proposed. Using a strain gauge based sensing device mounted on the robot gripper for vibration feedback, the orientation of the part relative to its path is controlled to reduce vibration. For the second problem, a sensor fusion system developed previously is used to provide feedback about the contact condition between two sheet metal parts. An integral contact controller (ICC) is used to correct any angular error between the parts to ensure full contact along the joint for subsequent welding. Experimental results confirmed the effectiveness of both control algorithms. The LEC reduced the vibration amplitude by up to 45%. The ICC reduced the angular error from 0.5/spl deg/ to 0.025/spl deg/ in 1.7 seconds.