An Empirical Analysis of Team Coordination Behaviors and Action Planning with Application to Human-Robot Teaming

Julie Shah, Cynthia Breazeal

Human Factors, The Journal of the Human Factors and Ergonomics Society, April 2010

Summary

                  Robots working in team with humans are increasing and the field of application is also covering stressed, highly uncertain, ambiguous and time pressured environments, therefore studies on team-working and human-robot cooperation are of deep interest. The approach in this paper is studying human-human interaction in order to apply the discovered rules in the design of a human-robot cooperation environment.

Studies on implicit and explicit communication affecting team working have been done, in fact it is demonstrated that implicit communication, including non-verbal cues improve team working in terms of efficiency. It also has been studied that team under pressure, uncertainty and complicated conditions can perform in the same way, if not even better, than teams not facing this kind of conditions.

There are already a certain number of HRI researches investigating robot capturing humans expression, gesture, there are systems capable also of processing human spoken orders and Fong et al. (2006) provided also the Human-Robot Interaction Operating System, which accomplished cooperation through a central task manager capable of decompose goals into high-level task assigns tasks either to a robot or a human.

Examples of implicit communication, which are used for improving team performances, are the use of periodic situation assessments, preplanning and dynamically redistributing workload among the team.

In implicit coordination the use of Shared Mental Models (SMM) is the main strategy working in the background, for example “cross-trained” team member share responsibilities and aspects which are capable of making the overall team more performing (Volpe, 1996), people tend through SMM to incorporate resources and capabilities of other team members into their own action planning. Stout et al. (1996) has identified 9 methods for enhancing SMM: 1) creating an open environment, 2) setting goals and awareness, 3) exchanging preferences and expectations, 4)clarifying roles and information to be created, 5) clarifying sequencing and timing, 6) discussing handling of unexpected events, 7) discussing how high workload affects performance, 8)pre-preparing information and 9) self-correcting. Stoud et al. (1999) and Orasanu (1990) found out that the most effective team tend to generate more type of planning behaviours.

The authors of the paper discuss the importance of “switching cost”, being an explanation for the benefits of implicit communication, meaning that the immediate response of a team member (caused by explicit communication) would cause degrade in team’s performance since a responding to the command may imply a waste of time in changing activity and this tends to be magnified in complex environments, provoking lack of flexibility and therefore efficiency.

Three hypothesis are taken: 0) team exhibit increased use of implicit coordination behaviour as time pressure increases, and coordination behaviour is positively correlated with improved team performance outcomes, 1) explicit communication will provoke immediate response, 2) explicit communication has higher specificity.

The experiment involve 30 couples, half of them working in a competition and time pressured environment; the task was regarding building 4 structures with toy bricks and, although the users could perform the 4 task simultaneously, not all the pieces for doing so where given, without letting them know.

Key Concepts

Human-Robot Cooperation, Team Working

Key Results

Hypothesis all were proven to be true. This study appears to be useful for robotics designers and suggests that robots should use explicit cues for an action that required immediate response, while efficient coordination should be promoted through implicit cues. If these principles are followed, communication will be more natural.

Trajectory Prediction for Moving Objects Using Artificial Neural Networks

Pierre Payeur, Hoang Le-Huy, Clement M.

Human Factors: The Journal of the Human Factors and Ergonomics Society

Summary

                  Prediction of objects trajectory and position is very important for industrial robots and servo systems, where this kind of information might be used for control, capture and other observation issues.

Parts’ trajectories are often unknown and catching object problem can be useful for example in a typical loading/unloading problem of a robot grasping an object (for instance s box) from a slide or a conveyor.

In such a condition of course accuracy is strictly important, both for achieving the goal and for safety purposes.

An optimum off-line method is path planning based on cubit polynomials with dynamics added (G. Sahar, Hollerbach, 1986), but this method involved a lot of computation and it is done off-line, being not flexible.

The authors propose the use of Artificial Neural Networks (ANN), which can be computed faster and most of all can learn the system, being adaptable and flexible. The problem can be defined as the prediction of the trajectory of a moving object in real time, with minimum error, with no collision on an arbitrary path.

·       Model

Time t₀ is defined as the time in which the manipulator starts moving, the prediction is done with the period of time T. The trajectory model is defined with a cubic function: χ(t)=1/6α₀t³+1/2β₀t²+γ₀t+χ₀, where χ represents the position, orientation, velocity or acceleration according on needs.

Neural network are trained to define the values of the coefficients α₀, β₀, γ₀ at each change of trajectory for which they have to be recalculated, the presence of pre- and post-conditioning allows the need of a long network so that in the end there are 3 inputs, two layers of 20 hidden cells each and one output cell.

The global predictive structure is a multiplication of this basic topology by the number of d.o.f. .The Network is modeled so that it basically obtains information about the change in position and not the exact position, this method ensures reduction of calculations, but pre and post conditioning are then of course required in order to ensure the start and continuation of the calculations. Sampling period T too many times also should be avoided, so the preconditioning module computes differences between the known positions and feeds the network is pseudo-displacement value, so the network will process it and generates the anticipated pseudo-variation of the position or orientation, the post-conditioning module finds then velocity and acceleration through kinematics. The processed data from the neural nets is normalized so that the maximum value is lower than the activation function maxima (in this case a bipolar sigmoidal activation function with a range from -1.0 to 1.0). Between two successive sampling there is a maximum coordinate variation ∆p_max, while limitation in resolution is the minimum detection possible (∆p_resolution), so that the number of possible steps to consider is S_Tsingle=2∆p_max/∆p_resolution+1, which must be globally should be powered by three for each of the inputs. The size of the data can be reduced since two successive position or orientation variation are always of a similar amplitude with a constant sampling rate, so it can be assumed hat there is a maximum difference, tables (Appendix C) help for calculations shown at page 151. Once the input triplet is processed with the cubic model already introduced (to ensure that the output data presented to the network are exact), these are normalized between ±0.8 to avoid extremes, at that point the back-propagation algorithm and the pseudo-random presentation order are used for training the data (triplets are not presented in a fixed order, but the entire set must be used within one epoch).

Key Concepts

Artificial Neural Network, Path Planning, Trajectory Prediction

Conclusion

The method appear to be optimal and working as the polynomial trajectory model but being more flexible. Anticipated values are also more reliable and variances are more similar between axis so that errors could be improved and controlled.

The Dexterous Workspace of Simple Manipulators

Zone-Chang Lai and Chia-Hsiang Meng

IEEE Journal of Robotics and Automation, Vol.4, No.1, February 1988

Summary

                  Many investigation have been one on the determination of extreme position of the end-effector of a manipulator, this paper is regarding instead studies on dexterous workspace with 6 joints axes, which results not been investigated due to complexity in computations.

A dexterous space is defined as a space in which a manipulator’s hand can rotate fully about al axes through any point.

In the manipulator’s workspace, the sphere, called “service sphere” ( SS(P)) , is constructed around a point P with the hand size of the manipulator as a the radius. When the manipulator’s hand reached pint P, then the position of its six joint (J₆) must be located on the sphere. The so called “service point” is the point on the sphere SS(P) which allows to access the position point P. The region containing more service point is called “service region”, if the manipulator can freely move in the region while keeping its hand in contact with P, then this point is defined as “free service region”.

O₄(H) is defined as the set of hand orientations with joints 4 to 6 able to rotate freely, so no boundary exists for a free service region corresponding to appoint in the dexterous workspace. It has been proven that the boundary of a free service region is described by either the boundary of W₁(4), which is the reachable space of joint 4 when joint from 1 to 3 are free to rotate, or the boundary of O₄(H).

In a dexterous space, the robot’s wrist should be capable of generating a full range of orientations, defining what is called “dexterous wrist”, so that when J₆ is at the boundary of a free service region, J₄ is at that boundary of W₁(4), so that it could be considered that the boundary of dexterous workspace is governed by the boundary of W₁(4).

Before performing computation hypothesis for the problem must be checked: it must be check is the robot’s wrist is dexterous, the condition that J₄ is not at the boundary of W₁(4) for all hand orientation must be determined. If a wrist has unlimited revolute joints and its twist angle of joints 4 an 5 equal ±90°, then conditions for the wrist being dexterous are verified.

The boundary of W₁(4) can be described or by rotational limits of joint 1 to 3 or by the dimensional constraint of the first three links. The equation which governs the boundary of W₁(4) can be obtained by solving det(G)=0, where G is the Jacobian matrix of the robot’s first three links. G is defined: [dx₄ dy₄ dz₄]^T=G[dθ₁ dθ₂ dθ₃]^T, where x₄, y₄ and z₄ are X, Y and Z coordinated of J₄.

Solving det(G)=0 then the function f₁(θ₁ θ₂ θ₃]=0 represents symbolically the W₁(4) boundary conditions.

The dexterous workspace may then be obtained by finding the hand position such that: f₁(θ₁ θ₂ θ₃]≠0 and θ_i1< θ₁< θ_i2, where i=1,2,3 are the joints.

For 6 d.o.f. robots this computation may be very difficult, the authors introduce 3 simple cases for commond practice: a PUMA type manipulator with an interesting orthogonal wrist, manipulator with a roll pitch-yaw wrist (nonintersting orthogonal wrist) and a manipulator with all unlimited joints and a roll pitch-yaw wrist and limited revolute joints.

Key Concepts

Dexterous Manipulators, Workspace Boundaries

Key Results

The PUMA type manipulator with an orthogonal wrist appears to have a spherical workspace, as already shown with different computation by Ruth. The manipulator with a roll pitch-yaw wrist (nonintersting orthogonal wrist) also appears to have spherical workspace, computations for both manipulator are reported at page 101 and 102. The manipulator with all unlimited joints and a roll pitch-yaw wrist and limited revolute joints has the difference that has to be described either by the dimensional constraint or by the rotating limit, since joint aren’t unlimited. 

Design of Tactile Sensing Systems for Dextrous Manipulators

Stephen C. Jacobsen, Ian D. MacCammon, Klaus B. Biggers, Richard P. Philipps

Control Systems Magazine, 1988

Summary

                  Tactile sensors appear to be extremely important for a certain amount of information they can deliver, although research on the design of tactile sensors is still on the way due to the simplicity in mechanics that grippers have. The main issue appear to be: understanding the ways which contact information can be used to control grasp and the development of the overall system itself. Technologies at the mere level of transducing appear to be mainly on hand, while problems appear to be on the higher level of organizing the overall system in which the sensors are applied. The overall mechanical manipulation system is schematically divided in 6 subsystems: command source, control, effector, observers, models and the physical environment.

The paper takes as an example the Utah/MIT Dextrous hand with tactile sensors applied on it, although it wasn’t intended for industrial application, the hand appears to be good for this kind of testing in order to verify speed, strength, range motion, capability for graceful behavior, reliability and economy. The authors propose a hierarchical structure of general requirements, for this at the first level there is transduction, which is at the most simple and contact level. At the second level there is preprocessing, which is strongly dependent on the transducer and influences reliability, size and mechanical behavior of a tactile sensing array. At a third level there is the multiplexing and transmission, where data is collected and forwarded to the following steps. At the forth level there is tactile data selection, in fact data must be filtered, since only part of the information is really interesting for practical purpose (as for vision sensors). At the fifth level there is tactile data interpretation, where information is mapped and sent to the sixth level, the multisensory fusion, where information is blended with the output of other sensory systems. At the seventh level there is the world model construction, where  multidimensional image is constructed from the data; finally at the eight level there is control of grasp and manipulation. At the transducing level the aim is to obtain a stable grasp, it is considered that at least 10 bits of force must be achievable. An important issue consist in data selection, mainly 3 methods are considered: Full Scan (good for small sensors and considering mainly all sensors activated continuously, being energy consuming and providing too much data if the system is too big), Reactive Scan (sensor work only in annotating a change in the system) and Anticipatory Scan (the sensors scan patches in different moment with high frequency).

Two main examples can be taken in the field of tactile systems: site-addressable sensing systems and line-addressable sensing systems. In the first case each sensor can be individually accessed via address and data lines. Computations lead to demonstrate that full scan be accomplished by each scan at a rate of f_l=35.5 Hz.

The second systems consist in sensors connected in an matrix, they addressed according to the row and column they are positioned, a similar system is in normal computer keyboards, which in fact are basic tactile sensors (with each key being a 1 d.o.f. sensor). In this case computations lead to demonstrate an operating frequency of 34.5 Hz, being little bit low, but justified by a system which a more efficient system which allows multiple sensors in the system.

Key Concepts

Tactile Sensors, Manipulators

Key Results

The authors decided 3 steps for implementation: designing a binary sensing network for many sensors connected, introducing proportional contact sensing network and in the end making a multi-parameter sensing system.

The system may include totally 2000 sensors, the bandwidth will therefore approximately be 1 MHz over a single data line, in the first step mechanical, electrical simplicity are searched together with virtually no delay of signal transmission. For the second phase magnets and Hall-effect sensors can be applied to achieve 6 dof, in this case the use of anticipatory scanning may be preferred. For the last step the concept of modularity becomes important for the integration of other sensor systems.