The method of the quasioptimal per energy e ciency design of the motion of triple-section anthropomorphic manipulator with seven degrees of mobility approximated in the form of sphere with hindrance in working area in a real time operation mode is provided in this article. Developed method is based on the iterative piecewise-line generation of the anthropomorphic manipulator motion path. This method has rather low computational complexity that allows working in the real time operation mode and gives it the exibility which adapts it for di erence commands. The task of quasioptimal for energy saving motion path of anthropomorphic manipulator the adaptation of iterative piecewise-line generation is method performed in order.
The required trends of robotics technology development are stipulated by the need to replace the human labour in similar operation performance or during work in potentially dangerous elds which can cause the risk to human health.
The subject of the research is limited by such a variety of robotic product as a manipulator. The manipulators for similar operations implementation are generally presented by the industrial manipulators with centralized power and program control. The industrial manipulators llfull the similar operations of installation, welding and painting. The power consumption during these operations is a very important question in the process of nal product manufacturing because it determines their self-cost. The methods e ective for the optimization of the target energy saving but cost-intensive in computational complexity can be applied while the manipulator is at work. The optimization of the manipulator movements is carried out only once and its results are used in the control program. The situation is di erent with the anthropomorphic manipulator which applied for work in the potentially dangerous condition for human like underwater, spaceward and radiation area. While performing rescue works, serving nuclear industry entity and accomplishing the technical maintenance service of technics in space the manipulator movement have low frequency. The operation value due to the energy e ciency is de ned by the amortized cost of electric power elements but not by the cost of electric power as against the industrial manipulators with centralized power network. These power elements have the limited amount of charge-discharge cycles. As well, their replacements connected with the complexity of isolated work areas demands extra costs. The hardware and software methods for decreasing energy consumption could be used under these conditions. The hardware-based are expressed by the usage of easer materials, engines with higher e ciency and etc. Hardware methods increase the manipulator cost in spite of their apparent advantages. Program methods use more e ciency algorithm of manipulator motion control. There are a lot of e cient but computationally complex methods of the motion path design. There are two approaches of the motion path design in literature. They are the approach based on the diagram theory and the approach of spline interpolation. The First approach is characterised by computational complexity and the second one is characterised by the complex choice of supporting points, the presence of redundant motions at few points and calculation complexity at many points.
The motion path design at redundant manipulators movement based on the evolutionary approach is considered in the researches [Kam07, Qi14, Qi14, Xid18].
Design methods of the manipulator movement based on the natural movement of a manipulator are presented in works [Liu16, Ren15].
The Numerical methods of design problems solution are presented in works [How14, Che17].
The Manipulator control under the above mentioned extreme conditions is realised in a real time operation mode. So the limitations, which keep out the use of the above mentioned algorithm of the motion path design to optimise the manipulator movement for energy saving are laid on the algorithm computational complexity. However this method do not allow to solve the task how to plan the motion path of the anthropomorphic manipulator with hindrance.
The aim of the research is to work out the motion path design method for the anthropomorphic manipulator on the basis of iterative piecewise-line generation in order to solve the task how to design the quasi-optimal, energy-e cient method in extensional space with hindrance at a real time operation mode. The Quasi-optimality of the suggested method is stipulated by the application of the \greedy method" that do not guarantee the global optimality of the above-mentioned method. 2
The kinematic diagram of the considered anthropomorphic manipulator with 7-degrees of mobility is shown in Figure 1(a). This kinematic diagram is used in robots of series AR-600, SAR-400, FEDOR manufactured by \AO NPO "Android Technique"" [NPO18]. The following indices are used in this Figure: B1 4 shoulder, elbow radoiocarpal key parts and working terminal, correspondently; A1 7 manipulator 7-degrees of mobility, A8 manipulator working terminal.
The design task is to nd set of intermediate positions for the manipulator movement in order to go from start to nal position while it rounds hindrance
Thus, the following initial data are required on the bases of the above mentioned task.
S - the numerical vector of the manipulator generalised coordinates in starting position ( index of s, i-type points at that it is generalised coordinate in starting position); B4;F { the coordinates of the working nal position B4 in the global Decartes coordinate system.
O{ the coordinates of hindrance center set in global Decartes System and its radius R.
Further the following variables are used.
IB1 B2; IB1 B2; IB1 B2{ the length of shoulder , elbow and hand elements of the manipulator. k the numerical vector of energy intensity coe cients for engines which rotation degree corresponds to generalised coordinates with relevant indices. These coe cients are required for energy consumption.
According to the above mentioned motion path design task the result of the applied method is to achieve the arranged set of the numerical vectors for the generalised coordinates describing the position that should be gone by the manipulator from the starting position in order to achieve aim point under condition of going round the hindrance. This arranged set is de ned as the path L = S ; 1; :::; F where, S { the numerical vector of the manipulator generalised coordinates in starting position, F { the numerical vector of the manipulator generalised coordinates in nal position. 3
Let's introduce the following notions used further in the article. Global Starting (GS) and global nal (GF) positions are starting and nal position of the motion under which the path design is accomplished. The inline motion paths are considered during the design of the manipulator motion path by the suggested method. Starting, intermediate and nal positions for every path are indicated as local starting (LS) local intermediate (LI) and local nal (LF) positions.
Let's use Denavite-Khartenberg presentation to develop the transform matrix between di erent system of coordinates connected with a manipulator elements. The Connected systems of coordinates are shown in Figure 1(b).
The rotation angles of the anthropomorphic manipulator connections are considered to be generalised coordinates to this manipulator of the circular type - . Developed method algorithm of movement path design shown on Figures 2(a) and 2(b). Let's describe every step.
Step 1. Input Data. As it was described earlier, the input data are the numerical vector of the manipulator generalised coordinates in starting position S ; the coordinates of the working nal position B4;F ; the coordinates of hindrance center O and radius R; length of elements IB1 B2; IB1 B2; IB1 B2; the numerical vector of energy intensity coe cients for engines K.
Step 2. Inclusion GS position. As the manipulator movement starts from GS position, the numerical vector of the manipulator generalised coordinates, rst of all, must be added to movement path L, which is presented as a sorted list.
Step 3. Calculation GF position. We must solve the kinematic inversed problem to control the manipulator as we know only the point of GF in which the e ector should be. We can use the approach suggested in [Pet18]. This approach allows to optimise the manipulator nal position thus, that at the in-line movement from start to nal position the power consumption is minimal, if the e ector reaches the target nal point.
It can be considered that engine energy consumption linearly depends on the change absolute value of the corresponding rotation angles at the manipulator connections while moving between two positions. Thus aimed function is equal to total energy consumption of all the engines while moving from start to nal position, and can be written as follows: f ( S ; F ; k) = kj F
S j (1) where F { the numerical vector of the generalised coordinates in nal position.
j F S j { elementwise diminution manipulation and module taking; The following limitation is put on the numerical vector of the generalised coordinates in nal position { e ector B4 is set to be in point B4;F : (a) The Kinematic Diagram
(b) Joint System of Coordinates where 0T8 { the transformation matrix from 8-th coordinate system in 0-th .
The method of generalized reduced gradient could be used for the purpose of this 7-dimensional optimization with three limitation-equality (2) task [Pet18].
Step 4. Recursive Algorithm Initialization. The following recursive algorithm allows to analyse the possibility of in-line motion in generalised coordinates between two positions. The intermediate position is introduced by \pushing out" the nearest motion path point to the center of the hindrance if the in-line motion is impossible because the path crosses the hindrance.
(a) The Scheme of the Movement Path Design Algorithm
(b) Recursive Algorithm Scheme
Thus, the following data as required to be sent during the rst iteration of the recursion if the recursive algorithm is initialized: generalised coordinates in starting S and nal F positions; logical type variable g, which informs us if the nal position is global or not (for this situation { yes); hindrance characteristics O and R; the numerical vector of energy coe cients k; length of elements IB1 B2; IB1 B3; IB3 B4; The motion path formed for the current period L (this motion path includes only the numerical vector of the generalised coordinates in the global starting position for the rst iteration)
Step 5. Recursive Algorithm. The reported positions are conceded as the local starting and the local nal inside the recursive algorithm. The inside input variable indicates if the local nal position is also the global nal. The following steps are performed at every iteration of recursive algorithm.
Step 5.1. Assay in-line Motion of element B1B2. The objective of the assay applied to the in-line movement for B1B2. is to nd the \worst" intermediate position of this element at the in-line movement from the local starting to the local nal position. If all the generalised coordinates depend on time linearly and reach the nal value simultaneously, the movement is in-line. The following function can describe this time dependence: where t { the normalized time changing from 0 to 1. 1 is complied to the time in a total scale required for the movement in the slowest connection. If a distance from the element B1B2 to the center of hindrance O is minimal in intermediate position, this position is called the \worst"
Step 5.2 Element B1B2 in-line Motion Capability Check. It is necessary to accomplish the following inequation to supply the possibility of the element B1B2 in-line motion:
The physical sence of this inequation is following: the time when the distance from the element B1B2 to the center of the hindrance is minimal the distance should increase the hindrance radius.
(t) = S + t ( F
S ) d1 (Y1 ( (t ))) > R (3) (4)
The assay of the in-line motion for the elements B2B3 and B3B4 as well as the check of their possibility in steps 5.3-5.6 are carried out if the inequation (4) is accomplished. This operations are similar to steps 5.1, 5.2.
The Local nal position is added to the motion path L on step 5.7 and the output from the current iteration of the recursive algorithm is performed if all the in-line motions are possible.
Thus the output from the recursion after the rst iteration will be performed if the in-line motion from the global starting position to the global nal position is possible.
If one of the element can not move directly transition to the step 5.8 introducing the intermediate position is performed.
Step 5.8. Intermediate Position Introduction. The intermediate Position is introduced if the worst position from 1 of the manipulator element to the center of the hindrance is less than its radius. To do this the manipulator is pushed out from the center of the hindrance with a some margin h subsequently starting the shoulder element.
If the in inquality is not performed is indicates that any element goes across the hindrance (4). The pushing out procedure of any manipulator element is the following.
Let's analyze any element BiBi+1 with known started Bi (xBi; yBi; zBi) and nal Bi+1 (xBi+1; yBi+1; zBi+1) points.
Let's introduce the new circle with the center in point O and with radius R + h where h is some magian that is necessary to protect the pushed-out manipulator touching the hindrance. Then, the element position in hand coordinates after pushing out will coincide with the tangency that is the closest to the circle started from the point Bi .
It is necessary to nd both of the tangency and the points of tangence K1 and K2 coordinates to nd the nearest tangency. These points lie in the crossing points of the sphere with (the center in point O and the radius R + h), the sphere with (the center in point Bi and the radius r = BiK1) and also the subspace crossing the sphere center and the element BiBi+1. The system of equations for this points is: 8 (x < : (x xO)2 + (y yO)2 + (z
zO)2 = (R + h)2; xBi)2 + (y yBi)2 + (z zBi)2 = r2;
N0x + N1y + N2z + D = 0; where, R2 = min hBiBi+12; BiO2 (R + h)2i; N = fN0; N1; N2g = BiO
BiBi+1 { the tangence vector to the drawing subspace. D = N0x N1y N2z.
To nd the nearest tangency to the element BiBi+1 we must choose those tangency which directional single vector has the bigger scalar module multiplication with the element vector: (5) (6) (7) After that, the segment of length BB should be laid out along the chosen tangency BiBi+1 from the point Bi: where, BiBi+10 { is the expected vector of the de ect position.
The following should be concided at the moment of elements pushing-out. After the shoulder element pushingout the change of hand coordinates of elbow and hand ends is happening as they depend on shoulder element position. Similar, after the elbow pushing-out it is necessary to recalculate the end of hand element coordinates.
For that in the algorithm after pushing-out every element, the recalculation of generalized coordinates, which describe its position, is performed. Further, the recalculation of hand coordinates for all the previous elements is performed with Denavit-Khartenberg presentation.
The values of the generalised coordinates in pushing-out position can be nd with help of the speci c solution with the help of inverse kinematic solution.
Let's de ne the value of the generalised coordinates for the manipulator in position indicated in Figure 1(b), as = f 1; 2; 3; 4; 5; 6; 7g :
The value of the generalised coordinates in pushing-up position P can be recalculated as: K = ( K1; if jBiK1BiBi+1j jBiK2BiBi+1j ;
BiK1 BiK2 K2; if jBiK1BiBi+1j < jBiK2BiBi+1j :
BiK1 BiK2 BiBi+10 =
BiKBiBi+1
BiK 3 = ataNn42=B(32TB4)22 x1;NB;32B22 y ; 0
r 2 arccos r+h 1
3 A + 15 ; where, N { number of operations; n { the coe cient depending on the de nite program realisation (for authors this coe cient is 42 103); r { the hindrance radius; h { the setting characteristic of the motion path algorithm; " { the setting precision of one-dimensional optimization process.
Thus, for the following example: n = 42 103; " = 0; 01; r = 50cm; h = 1cm , the number of operations for the worst case is N 1; 1 106: More than, the design of the motion path for the anthropomorphic manipulator can be done so that the motion can start before the total calculations are completed.
Considering the fact that modern central and graphic processors have the capacity of about 1011 FLOPS, the designed method can be applied in a real time operation mode.
In order to compare the designed method with other well-known methods used design the path for the anthropomorphic manipulator, let's compare computational complexity.
Neural network and graphical analytic methods are based on a ow chart. The implementation of Voronoi diagrams [Qi14] is improved variant of graphical analytic methods. According to the data published in [Psh15], the complexity class of the methods based on the compilation of ow charts is O n2 , and on the Voronoi diagram is O (nlg (n)), where, n the number of elements on the frame.
On the basis of the research we can build the diagram including the number of required operations depending on the resolution power N within constant factor. The results of the numerical simulation are shown in Figures 3(a) and 3(b).
As the diagrams show the number of operations even for the resolution power N = 100 (positional accuracy in an engine is about 1 ) go beyond the accepted limits in a real time operation mode. (8) (9) (a) The Number of the Operations for the Meth- (b) The Number of the Operations for the Methods based on the Voronoi Diagram ods based on the Road Map.
In order to prove this, the further practical implementation of the algorithm considering the abovementioned recommendations be performed. Energy consumption of the anthropomorphic robot manipulator when performing the target operation in the working area with a typical obstacle was reduced by 11.2% without the use of an intermediate solution to the optimization problem of calculating the generalized coordinates in the nal position, and by 16.6% with its use. The e ectiveness of the proposed solution to the optimization problem of nding generalized coordinates in the nal position as a whole, and the proposed objective function in particular indicates by saving of energy consumption on 38%.
It's also planned to measure the operation time and analyze the energy e ciency of the methods during the computing experiment. 4
The aim of the article was to work out the design method of motion path for the anthropomorphic manipulator on the basis of the iterative piecewise-line generation in order to solve the design problem of the quasioptimal energy e cient path in extensional space with a hindrance in a real time operation mode.
The algorithm of the adopted method as well as all the required calculations are presented in the article. The designed method allows to round the hindrance { approximated by the sphere and the applied \greedy method" makes it possible to achieve the quasioptimal per energy e ciency path. The computational complexity using this method is equal to 1; 1 106operations. Modern central and graphic processors have the productivity of about 10 FLOPS. So, the developed method can be applied for designing the motion path of the anthropomorphic manipulator in areal time operation mode.
As the design is considered to be quasioptimal and based on the \greedy method", the software implementation of this method and its comparison to the other methods of energy e cient path design for the anthropomorphic manipulators per the criterion of energy e ciency are to be urgent. 5
The description of the designed methods of quasioptimal per energy e ciency motion path of the anthropomorphic manipulator in a real time operation mode is presented in the article. The methods is based on the numerical approach of the iterative piecewise-line generation of the point motion path function in the space with hindrances and its adaptation to the anthropomorphic manipulator in a view of the design method of the optimal path function in a work area with hindrances. The solution method of the kinematic inversed problem for the triple-section anthropomorphic manipulator with 7-degrees of mobility on the basis of the Denavit-Khartenberg presentation and the solution of the nonlinear optimization task per the criterion of the energy e ciency by the numerical method of the generalized reduced gradient were used to build the starting path. In order to move between the positions of the anthropomorphic manipulator, the formulas of the inversed kinematic approach adopted to this task were developed. The initial data for the suggested methods are start, intermediate and nal positions of the manipulator motion path as well as the generalized coordinates used to ful ll established destination operation. The analysis of the computational complexity showing the possibility to perform the mentioned operations in a real time operation mode was conducted.
The software implementation of the developed method and its comparison to the other energy e cient path design methods of the anthropomorphic manipulator motion are considered to be the further tasks. 6
The research is accomplished in the framework of the scienti c project \The Development of Hardware and Software System Control Complex per the Solution of the Dynamic and Kinematic Inversed Problem" in the framework of FZPIR 2014-2020 (unique ID-RFMEFI57517X0166) and with the nancial aid of the Russian Ministry of Education and Science. [Pet16]