Tuesday July 9
Full Day (8:30am – 5:30pm)
Organizer: John T. Wen (Rensselaer Polytechnic Institute), Abhijit Chakraborty (United Technology Research Center)
This workshop will focus on sensing and control challenges and opportunities in the fast-changing landscape of robots in manufacturing. A panel of researcher from industry and academia will present a set of case studies on the effective integration of robots into a manufacturing environment to assist human workers to increase performance and productivity. The examples will draw from application domains such as aerospace, renewable energy, and automotive industries. The underlying control problems will include vision and force guided motion, collaborative human-robot assembly, robotic manipulation of deformable materials, robot grasping and dexterous manipulation, learning control, and others. Additional discussion will include distributed control and communication middleware architecture and the advantages and challenges of open source software community. The workshop will highlight the recently established Advanced Robotics for Manufacturing (ARM) Institute. As part of the Manufacturing USA Network, ARM is funded by the Department of Defense to address these challenges by bringing research results the factory floor.
The workshop would be of interest to researchers with general control background who are interested in learning about industrial applications and use-cases related to robot-assisted manufacturing and smart factory, and practitioners who would like to learn about the latest research tools and results related to collaborative human-robot control systems.
Schedule
8:30-8:45 
Introduction and Overview (John Wen, Rensselaer Polytechnic Institute)
8:45-9:20 Robot Assisted Manufacturing (Harry Asada, Massachusetts Institute of Technology)
This talk will discuss two current projects related to robot assisted manufacturing: Robot-Assisted Wire harness Installation (funded by ARM) and wearable extra limbs for ergonomic accommodation of aircraft assembly workers (funded by Boeing).
9:20-9:55 Robotic Assistants for Composite Prepreg Sheet Layup (SK Gupta, University of Southern California)
Automated composite prepreg sheet layup require the use of multiple collaborating robots to operate under human supervision. To be useful in this application, robots will need to act as smart assistants by (1) programming themselves, (2) efficiently learning from the observed performance, (3) safely operating in the presence of uncertainty, (4) appropriately calling for help during the execution of challenging tasks, and (5) effectively communicating with humans. This presentation will provide an overview of the advances in artificial intelligence that are being used to enable robots to automatically make decisions to meet aforementioned requirements in composite prepreg sheet layup application. First, we will present an approach for automatically generating trajectories to enable robots to program themselves from high-level human instructions. Second, we will describe self-directed learning methods to equip robots with the ability to learn from observing the performance of previously-executed tasks and adapting their plans. Third, we will describe methods for robots to operate safely in the presence of uncertainty by accounting for potential contingencies. Fourth, we will discuss computational methods for endowing robots with introspective capabilities so that they can seek help from humans. Finally, we will present context-aware reasoning for enabling robots to elicit human guidance during the decision-making process.
9:55-10:15 Break
10:15-10:50 Motion and force control in industrial processes (Mike Klecka, Abhijit Chakraborty, United Technology Research Center)
Material processing in aerospace industries often involves contacts between tools and parts. This talk will discuss using robot motion and force control to automate these processes.
10:50-11:25 Vision and Force Guided Robot-Assisted Assembly of Large Structures (John Wen, Rensselaer Polytechnic Institute, Shridhar Nath, GE Global Research)
This talk will present the control architecture and result of robotic assembly of a large segmented structure. The aim of the project is to demonstrate the potential of robotic technology to reduce cycle time, enhance assembly quality, and improve worker ergonomics, as compared to the current manual or fixture-based approaches. Our approach involves the ROS-based integration of an ABB IRB-6640 industrial robot, human operator input, stationary and robot mounted cameras, and force/torque sensor. The key components of the system include: human directed path planning, sensor-guided robot motion, and simulation preview using the same run-time code with virtual sensors. Point-to-point motion planning is based on local optimization guided by operator specified way points, subject to the collision avoidance, joint limits, and velocity and acceleration constraints. Contact tasks such as panel pick-up and placement are based on compliance control. The pick-up and placement locations are determined by vision and tags on the panels and assembly nest. The required millimeter level alignment accuracy is achieved by image-based visual servoing. Robot motion is commanded through the ABB External Guided Motion (EGM) mode which updates the joint set point at the 4ms rate. EGM itself also exhibits nonlinear dynamics. We will also present results using learning control to compensate for the dynamics. The training is performed based on the ABB dynamic simulator, RobotStudio, and the application to the physical robot applies the transfer learning technique.
11:25-12:00 Evaluation of Mobile Robot SLAM and Navigation for Long-Term Autonomy in Manufacturing Environments (Jasprit Singh Gill, Mark Tomaszewski, Pierluigi Pisu, Yunyi Jia, Venkat Krovi, Clemson University)
In recent times, a number of reference implementations of Simultaneous Localization and Mapping (SLAM) and navigation techniques have been made publicly available via the ROS Community. Several implementations have transitioned to commercial products (vacuum robots, drones, warehouse robots, etc.). However, in such cases, they are specialized and optimized for their specific domains of deployment. In particular, their success criteria are based primarily on mission completion and safety of humans around them. In this light, deployment in any new operational design domain requires a careful verification of the performance and re-optimization of the technology. We seek the technological gaps that need to be addressed to ensure that the mobile robots are fit for automotive manufacturing environments. Automotive final assembly environments pose significant additional challenges for mobile robot deployment. First, they consist of tasks that involve manipulative and decision-making skills that humans excel at, necessitating robot operation in collaboration with humans. Second, they are replete with relatively unstructured tasks with significant uncertainty and dynamic elements, requiring the robots to be robust enough to changing environments and adaptive to flexible mission goals. Third, due to sequencing of the assembly processes and their impact on the overall production line, the tasks involved are time sensitive. This requires the robots to perform consistently even with the lack of structure and uncertainty. Thus far, the focus of autonomous mobile system researchers has been primarily on developing the functionality and optimizing its performance. Hence, the popular metrics for evaluating such systems are based on accuracy, quality and resource consumption. However, to evaluate them for long-term operation in manufacturing environments, these metrics are insufficient. This research evaluates the state-of-the-art SLAM and navigation implementations based on existing metrics for robustness & consistency, and where applicable, suggests a more apt metric. The evaluation uses ASTM F3244-17 standard test method. It is performed on a simulated robot in Gazebo environment, Clearpath OTTO1500 as well as Ridgeback robots in an industrial testbed environment.
Automotive Final Assembly: A New Playground for Human-Robot Collaboration (Jasprit Singh Gill, Yi Chen, Farbod Akhavan Niaki, Mark Tomaszewski, Weitian Wang, Laine Mears, Pierluigi Pisu, Yunyi Jia, Venkat Krovi, Clemson University)
The global trends towards user-customized products has led to a significant increase in the number of product configurations / variants that ultimately need to be manufactured. Consequently, the manufacturing production systems have evolved to support the realization of high-volume high-mix (i.e. build to order) product portfolios by adoption of advanced automation paradigms. The automotive assembly plants are no exception. As an example, in BMW plants, it is estimated that customer selection of options can create 1013possible unique product combinations for their vehicles.
Over the decades, the automotive industry has seen significant incorporation of traditional automation and industrial robots in the low variation part of the manufacturing process such as body shop (e.g. to align and weld together various metal parts to create the body-in-white) and paint shop (e.g. to allow for high quality consistent painting). Fixed base manipulation robots dominate the deployment landscape working either in exclusive fenced zones or with a limited region for collaboration with humans.
However, automotive final assembly has been one arena that has not witnessed a significant adoption of robotics and automation due to inherent challenges. First, significant portion of the customization occurs in final assembly and hence disproportionally contributes to the variability (compared to the body shop). Human assembly line workers (referred to as associates in this chapter) now need to: (1) Understand the operation to be carried out for a specific vehicle; (2) Pick the appropriate parts; (3) Assemble them on the vehicle using a variety of tools and processes (e.g. bolting, gluing, clipping); and (4) Conduct quality checks on the operation. This exposes the associates to two major loads: (1) Physical load: avoid repetitive strain injuries from handling (heavy) parts or conducting ergonomically challenging operations (e.g. overhead assembly); and (2) Cognitive load: understand which specific part and assembly operation is required for the next vehicle. Secondly, each vehicle spends only a few 10s of seconds at each assembly station making the assembly operations time sensitive. Thirdly, multiple dynamic elements like moving conveyors, stray carts, and human co-workers walking in and out to collect tools and parts to be assembled, contribute to a lack of structure at the auto-assembly station.
The Smart Companion Robot (SCR) project aims to demonstrate the viability of an intelligent mobile manipulator robotic system to assist and augment human associates in automotive final assembly. The physical assist is provisioned in terms of helping transport medium-heavy parts from subassembly areas to reduce worker fatigue/repetitive injuries. The focus for coupling sensor intelligence to the mobile manipulator platform is directed along three tasks: Task 1: Situational awareness for up-to-date monitoring of obstacles and autonomous base navigation for obstacle avoidance; Task 2: autonomous arm motion planning for manipulation in cluttered environments; and Task 3: digital-twin /process-simulation for what-if analyses of human-robot cooperation scenarios. The overall build-system and operator orchestrate the process steps using shared and supervisory control.
12:00-1:15 Lunch Break
1:15-1:50 Trust-based Control, Decision-Making, and Motion Planning for Human-Robot Collaboration Systems (Yue Wang, Clemson University)
Robots and autonomous systems are becoming an essential component that empowers economy and human possibility. The talk will begin with an overview of human-robot collaboration systems and human trust in robots. Next, we will introduce computational trust modeling. We then present our works in trust based haptic tele-autonomous operation of mobile robots, trust-based human-robot collaboration in manufacturing, trust-based symbolic motion planning for multi-robot systems, and trust-based information management for vehicle platooning under cyber attacks. The talk will conclude with a discussion about future research.
1:50-2:25 Safety and Security in Cyber Physical Production Systems (Azfar Khalid, Nottingham Trent University)
Development of future manufacturing systems needs to address flexibility, mass customization, intelligence and context-based learning to produce smart products. These systems are characterized through networked, cooperating objects and are able to capture, store, process, and communicate data. They are called cyber-physical production systems and forms the basis of novel interaction and collaboration opportunities between devices, manufacturing machinery, raw materials, working robots, humans and the plant environment. In this context, human-robot collaboration is an area for the future factory floor common applications. Due to the diverse capabilities of CPS and the fact that human robot collaborative systems have to address and fulfil industrial safety requirements to a major extent, there is a particular need to systematically guide the development of such cyber physical production systems. We consider that in such a system, human worker participation is integrated in a safe and secure connected environment. Based upon a real-world use case derived from automotive industrial assembly, the outcomes are applicable in order to develop safe human-robot collaboration in cyber physical production systems. The system is also studied from security standpoint in that cyber-attacks are categorized according to the extent on controllability and the possible effects on the performance and efficiency of such CPS, resulting in the development of a consolidated mitigation plan.
2:25-3:00 Advanced Robotics for Manufacturing (ARM) Institute Technology Roadmap (Arnie Kravitz, Chief Technology Officer, ARM Institute)
This talk will present challenges and opportunities in industrial robotics based on the ARM Institute Technology Roadmap. The Roadmap is generated based on extensive ARM membership discussion and input.
3:00-3:15 Break
3:15-4:45 Panel Discussion