An object collision detection apparatus.

Field of the Invention

This invention relates to an object collision detection apparatus that determines in advance, using computer simulation techniques, whether two objects will collide, or detects likewise the location and the time of collision, and more particularly to one that is designed to attain sufficient accuracy even at a high speed.

The invention can be applied to control of a robot, a numerically controlled machine, and other automated production machines.

In the following, the term "collision" means that a moving object hits another object, while the term "interference" means a state in which two objects overlap each other.

Prior Art

In a planning stage of production, it is recognized that geometric simulation using shape models is effective in verifying the motions of a robot or working process of an NC machine. In such simulation, it is important to detect collision between objects. For example, in the so-called off-line robot motion teaching, in which the motion is taught to the robot without stopping the production line, it is essential to check collision between a robot and another robot or object in advance, by simulation.

Conventional techniques for detecting collision can be roughly divided into the following three types.

Sampling Method

This technique determines the positions and orientations of moving objects at sufficiently short time intervals delta t and solves a problem of statical interference at each sampling time. Refer to "A Method for Checking Interference at High Speed in Off-line Robot Teach­ing" (in Japanese), Ozawa et al, Journal of Robotics Society, Vol 4, No 4, pp 5 - 13, April, 1986 for details.

The sampling technique is widely employed because it is easy to implement. However, if the time interval delta t for sampling is too large, there is a possibility of overlooking collision. On the other hand, if delta t is too short, the amount of calculation required increases.

Envelope Generation Method

This technique generates volumes swept by moving objects and considers them as solid bodies, and then investigate interference between the solid bodies. In this envelope generation technique, it is still difficult to generate the envelope. In addition, although the occurrence of the collision can be detected, there remains a problem that the time and status, when and where the collision occurs cannot be obtained.

Intersection Calculation Method

This technique mathematically expresses the trajectories of sur­faces, edges and vertices of moving objects as a function of time, and solves equations between them to find intersection points. Refer to "Interference Detection Among Solids and Surfaces" by J W Boyse, Commu­nication of the ACM, Vol 22, No 1, pp 3 - 9, 1979 for details of this technique.

In the intersection calculation method, equations can be analyti­cally solved in case of pure rotation or translation. However, in case where trajectpries of general motions are expressed using three order of functions of time, the equations become of sixth or ninth order of time. The amount of calculation increases because convergence calculation is needed to solve them.

In any case, a tremendous amount of calculation is a problem that is always pointed out in determining collision between moving objects by using a three-dimensional geometric models. The causes of this are considered complexities of the shape and trajectories of objects. The complexity of the shape has been in the past considered. The simplest method of reducing the amount of calculation is rough check by using a rough shape containing the exact shape such as an envelope sphere, an envelope rectangular parallelepiped or an envelope cylinder. This utilizes a condition in which collision does not occur; "Collision does not occur if envelope shapes do not collide," which is the contraposi­tion of a necessary condition in which collision occurs; "Envelope shapes collide with each other if collision occurs."

The effectiveness of this approach is well known. However, supp­ressing increase of the amount of calculation from another aspect of the complexity of trajectories has not been considered in the methods proposed. Thus, the invention provides a technique for suppressing the increase in the amount of calculation by considering trajectories.

Published Unexamined Japan Patent Applications 58-22690 and 60-99591 are related to the present invention. Both employ the sampling method, and do not consider trajectories as does the present invention.

An object of the present invention is to provide an object colli­sion detection apparatus that can detect collision with no overlooking and reduced amount of calculation, and that can also detect the time and location of collision.

Summary of the Invention

According to the present invention, the principle is considered that two objects do not collide if the condition that the minimum distance between two objects at the start or the end of movement is larger than the sum of the "maximum movement lengths" of the two objects is satisfied. Further, if the condition is not satisfied, the trajec­tory is divided, and shorter trajectories obtained through the division are again checked for the above condition. Such a procedure is recur­sively executed to detect any collision without fail. In a case where the minimum distance between the two object becomes less than a pre­determined threshold, and they are approximated to collide with each other, the processing is discontinued, so that the recursive procedure is not executed infinitely.

The "maximum movement length" means the longest of the lengths of paths on which points on the surface of an object move respectively. In order that the invention may be fully understood a preferred embodiment thereof will now be described with reference to the accompanying draw­ings in which: <ul id="ul0001" list-style="none"><li>Figure 1 shows a system configuration of an embodiment of the present invention.</li><li>Figure 2 shows a diagram illustrating a principle of the present invention.</li><li>Figure 3 shows a flowchart illustrating an algorithm employed in the embodiment of Figure 1.</li><li>Figures 4, 5 and 6A through 6E show diagrams illustrating the flowchart of Figure 3.</li><li>Figure 7 shows a diagram illustrating a method for calculating movement length in Figure 3.</li><li>Figure 8 shows a diagram for comparing the amount of the calcula­tion of the algorithm of Figure 3 with the sampling method.</li></ul>

Referring now more particularly to the drawings.

Figure 1 shows a system configuration employed by an embodiment of the invention. In the figure, the system consists of an IBM 5080 graphic system 1 and an IBM RT-PC personal computer 2. The personal computer 2 constitutes a data base 3 and collision detector section 4, which is implemented in software. In the figure, various functions are represented by blocks to help understand the configuration of the collision detector section 4. Of course, it is possible to realize each block with hardware.

A minimum distance calculating section 5, a maximum movement length calculating section 6, a collision presence determining section 7, and an end condition determining section 8 correspond to steps S1, S3, S4 and S2 in Figure 3, respectively. The operations in the system will become clear from the following detailed description.

Conditions of judgment on collision

First, a necessary condition for causing collision between moving objects is described, and then a condition in which collision will not occur is derived based on that condition.

As shown in Figure 2, objects A and B move at time ts &lt; = t &lt; = te, and notations are as follows: dmin(ts,A,B) : Minimum distance between objects at time ts dmin(te,A,B) : Minimum distance between objects at time te lmax(ts,te,A): Maximum movement length of object A during ts &lt; = t &lt; = te lmax(ts,te,B): Maximum movement length of object B during ts &lt; = &lt; t = te where the "maximum movement length" means, as described in the above,the longest of lengths of paths on which points on the surface of an object move.

When objects A and B collide at time ts &lt; = t &lt; = te, the following two conditions are always established: Condition P dmin(ts,A,B) &lt; = lmax(ts,te,A) + lmax(ts,te,B)

"The sum of the maximum movement lengths of the objects A and B in the time interval of ts &lt; = t &lt; = te is larger than the minimum distance at the movement start time ts." Condition Q dmin(te,A,B) &lt; = lmax(ts,te,A) + lmax(ts,te,B)

"The sum of the maximum movement length of objects A and B in the time interval of ts &lt; = t &lt; = te is larger than the minimum distance at the movement end time te." That is, (Collision occurs) -£ (P Q) (P Q) is a necessary condition to cause collision. By taking contraposition of this proposition, a suffi­cient condition in which collision will not occur is derived. <o ostyle="single">(P Q)</o> -£ <o ostyle="single">(Collision occurs)</o> (P Q) -£ (Collision does not occur) Namely, the following judging condition is obtained. [Determining condition of collision] If dmin(ts,A,B) £ lmax(ts,te,A) + lmax(ts,te,B) or dmin(te,A,B) £ lmax(ts,te,A) + lmax(ts,te,B) is established, collision will not occur at ts &lt; = t &lt; = te.

Collision detecting algorithm

Here, collision is defined as a state in which two objects approach each other within a sufficiently short distance Dcol.

The most important points of the collision detecting algorithm of the present invention are to judge the determining condition derived in E1, and to recursively divide the trajectory into two parts.

Now, a case where two moving objects A and B move independently at time ts &lt; = t &lt; = te in a three dimensional space, changing their orientations, will be studied. First, a function STATUS (ts, te) is defined for any time ts &lt; = t &lt; = te for investigating a status on collision according to the flowchart of Figure 3. The function returns one of four statuses of Figure 4 as a result value. <ul id="ul0002" list-style="none"><li>Status A: Collision is detected at t = te.</li><li>Status B: Collision can not occur at ts &lt;= t &lt; = te.</li><li>Status C: Collision is possible at ts &lt; = t &lt;= te.</li><li>Status D: Collision shall occur at ts &lt; = t &lt; = te.</li></ul>

Namely, as shown in Figure 3, dmin(ts,A,B) and dmin(te,A,B) are calculated, while interference between objects A and B at time t = te is checked (S1). This checking may be secondarily performed in finding dmin(te,A,B) from shape data or position data of objects A and B. Then, if it is recognized in the step S2 that there is interference at t = te, STATUS = D is outputted. If it is not recognized that there is inter­ference, the checking goes to the step S3. In the step S3, lmax(ts,te,A) and lmax(ts,te,B) are calculated, and then judgment is made on whether the above-mentioned determining condition on collision is established (S4). If the determining condition is not be establish­ed, STATUS = C is outputted.

If the determining condition is established, existence or non-­existence of approximate collision is judged in the step S5. If there is no approximate collision, STATUS = B is outputted. If there is approximate collision, STATUS = A is outputted.

In a case where STATUS = C or D is outputted, the trajectory is divided into two parts in the step 6, and a procedure COLLISION is recursively invoked for each trajectory part. In a case where STATUS = B is outputted, whether or not there is a remaining trajectory part, is judged in the step S7. If there is, the procedure COLLISION is invoked for the next part in the step S8. If there is no remaining trajectory part, the processing terminates. It also terminates if STATUS = A is outputted (S9).

When function STATUS(ts,te) is used, it is possible to find exis­tence of collision at Ts &lt; = t &lt; = Te, and the time when the collision occurs by invoking the procedure COLLISION(ts,te) with ts = Ts and te = TE. The following procedure COLLISION(Ts,te) is written in a PASCAL like language.<img file="EP0307091A2_D0001.tif" />

In a case where the result of STATUS(ts,te) is status C or D, there remains a possibility that collision will occur at time ts &lt; = t &lt; = te, so that the time interval is divided into two parts, ts &lt; = t &lt; = (ts+te)/2 and (ts+te)/2 &lt; = t &lt; = te, each of which time interval is recursively investigated. The processing of COLLISION(Ts, Te) is completed when STATUS(Ts,Te) returns to status A at a certain time, or status B in the whole time interval.

In the flowchart of Figure 3, judging dmin(te,A,B) &lt; Dcol is to prevent the procedure from being infinitely executed if there is colli­sion at Ts &lt; = t &lt; = Te. That is, a period including the collision is divided, one of the divided shorter periods includes the collision, and thus the division is infinitely performed. dmin(te,A,B) &lt; Dcol is a condition used to terminate the division. If Dcol is made smaller, accuracy is improved, while the amount of calculation increases. If Dcol is made larger, accuracy is lowered, while the amount of calculation decreases. However, lowering of accuracy means that collision is determined to be likely to occur, even if collision does not occur, and thus no collision is overlooked at all in this case. This differs from the above-mentioned sampling method.

Multiple collision may occur during a continuous motion. In such a case, the collision that occurs first can be detected by always process­ing two divided time interval parts in order of the first half and then the second half.

In a case as shown in Figure 5, the actual processing flow is as follows wherein Ts = 0 and Te = 1: 0 &lt; = t &lt; = 1 (Figure 6A) STATUS = C There is no possibility of collision; Divides into two parts of 0 &lt; = t &lt; = 1/2 and 1/2 &lt; = t &lt; = 1 0 &lt;= t &lt;= 1/2 (Figure 6B) STATUS = B There is no possibility of collision. 1/2 &lt; = t &lt; = 1 (Figure 6C) STATUS = C There is a possibility of collision; Divides into two parts of 1/2 &lt; = t &lt; = 3/4 and 3/4 &lt; = t &lt; = 1 1/2 &lt; = t &lt; = 3/4 (Figure 6D) STATUS = D Collision shall occur; Divides into two of 1/2 &lt; = t &lt; = 5/8 and 5/8 &lt; = t &lt; =1 1/2 &lt; - t &lt; - 5/8 (Figure 6E) STATUS = A Collision is detected at t = 5/8 (approximately).

Minimum Distance and Maximum Movement Length

Finding out the minimum distance between objects and the maximum movement length of an object is a major computation in the present invention, and its method is described. It is assumed that the shapes of the objects are represented as polyhedrons.

The minimum distance between two polyhedrons can be obtained by calculating the distance of each pair of edges, and the distance of each pair of a vertex and a face, and then comparing them. Assuming that the numbers of faces, edges and vertices of objects A and B are Fa, Ea, Va, Fb, Ea, and Vb, respectively, the number of their combinations are N = FaVb + VaFb + EaEb. Thus, the amount of calculation increases in the order of the square of the number of geometric elements. The increase of the amount of calculation due to the complexity of the shape is a common problem to the calculation using a three-dimentional model. In this regard, the number of the combinations is reduced by using simpler approximate shapes, or by considering that it is sufficient to find out sets of opposing faces and to make edges and vertices contained within them the subject of calculation, so that the efficiency of the calcula­tion can be improved.

Next, the description is made for a method of calculating the maximum movement length. The one with the largest movement length among points on an object represented by a polyhedron is one of the vertices of the polyhedron. Therefore, it is sufficient to calculate movement lengths only for vertices, and to compare them. If trajectories of vertices are given in functions of time X (t), Y (t) and Z (t), the movement length l is expressed by the following equation:<maths id="math0001" num=""><img file="EP0307091A2_D0002.tif" /></maths>

For simple movement as in the following, the indefinite integral has a solution. Translation at a constant speed: X(t) = Axt + Bx, Y(t) = Ayt + By, Z(t) = Azt + Bz l = Ax + Ay + Az (Te - Ts) Rotation at constant angular velocity: X(t) = R sin( t), Y(t) = R cos( t), Z(t) = A l = R (Te - Ts)

However, a more generic method for representing a trajectory is required to deal with complicated movement of an object such as a robot. One possible method is to approximate the trajectory of a vertex with spline curve as following equations with time t as a parameter: X(t) = Axt + Bxt + Cxt + Dx Y(t) = Axt + Bxt + Cyt + Dy Z(t) = Axt + Bzt + Czt + Dz

In this case, the movement length of the vertex is:<maths id="math0002" num=""><img file="EP0307091A2_D0003.tif" /></maths> Because this indefinite integral cannot be generally solved, a numerical integration is used. However, if, as shown in Figure 7, an trajectory is defined as a Bezier curve by control points P, it is possible to use the length of a polygonal line connecting the control points P, lcontrol, instead of the movement length l in the determining a condition of collision because the length of the polygonal line lcontrol is always longer than the curve l. lcontrol can be calculated by an amount of calculation less than that for l.

Evaluation of the Algorithm

The following three points are important as conditions necessary for an algorithm for detecting collision between moving objects. . There is no overlooking of collision. . Time and state where collision will occur can be detected. . The amount of calculation is small.

An algorithm with a small amount of calculation is required, particularly for an interactive system such as an off-line teaching system of a robot. Now, the amount of calculation in this algorithm is discussed by comparing with the sampling method that has been widely used.

Because much calculation is required for, not limited to the detection of collision, a problem in which a three-dimensional shape and movement of six degrees of freedom in a space are processed by a com­puter, an algorithm, satisfying the following two points is desirable, to reduce the required calculation to a practical amount. . Amount of calculation corresponds to the "accuracy of solution." . Amount of calculation corresponds to the "possibility of colli­sion."

Here, the "possibility of collision" is intuitively considered as follows: Low possibility : "Objects that make a relatively small movement with each other at remote positions." High possibility: "Objects that make a relatively small movement while keeping a close positional relation."

First, the "accuracy of solution" is considered. In thet sampling method, the "accuracy of solution" depends on the sampling time interval delta t. Although it is sufficient to make delta t smaller to obtain better accuracy, it increases the amount of the calculation. Although the amount of the calculation may be reduced by making delta t larger in case where so high accuracy is not required, the possibility of over­looking collision arises.

On the other hand, in this algorithm, it is assured that there is no overlooking, regardless of the "accuracy of solution." Because, as described above, when objects become close each other within distance Dcol, it is detected as collision, the value of Dcol itself may be considered as the "accuracy of solution." As in the sampling method, the amount of the calculation increases as Dcol is set to be small, while the amount of the calculation decreases as Dcol is set to be large. However, collision is never overlooked even if Dcol increases and the "accuracy of solution" is made worse.

Next, consideration is made on the relation between the "possibili­ty of collision" and the amount of the calculation. Figure 8 shows the results of a comparison between the sampling method and this algorithm. Here, the numbers of repetitions of unit processing, I and L are defined as follows: I: Number of repetitions of "checking of interference at the steady state between the objects" that is a unit process in the sampling method L: Number of repetitions of "calculation of the minimum distance between the objects and the maximum movement length" that is a unit process in the algorithm

Assuming that calculation time required for units processes are Ti and T1, the amount of the calculation as a whole should be estimated as Ti I, and T1 L. However, here, I and L are simply used as measurements of the overall amount of calculation. This is because, although T1 is generally larger than Ti, they can be considered the same in regards to their order.

Now, it is assumed that in the sampling method the number of divisions of a trajectory providing sufficient accuracy without over­looking is n. Because the trajectory should be completely checked for an example in which no collision occurs, always I = n regardless of the "possibility of collision." For an example in which collision occurs, the processing is completed at the time of detection, so that I varies in a range of 1 &lt; = I &lt; = n and I = 1/2 as an average. However, in this case, the value of I does not vary in relation to the "possibility of collision."

On the other hand, in the algorithm, the value of L corresponds to the "possibility of collision." For examples with high possibility of collision such as (C) or (C′) of Figure 8, the required amount of calculation becomes L = n, the same number of division as in the sam­pling method in the worst case. However, for examples such as (A) or (A′) with low possibility of collision, the processing is completed at L = 1 in a minimum case. Generally, L in an average case cannot be estimated in a simple manner because it depends on combinations of multiple factors such as the positional relation of trajectories, the shapes of the objects, and the state where collision occurs. However, for an example with average difficulty, it is possible to estimate L = log (n) when it is assumed that the division of an trajectory into two parts occurs at with a probability of 1/2. As an example which has less possibility of collision, division of an trajectory occurs fewer times, so that L becomes smaller. Thus, this algorithm is superior to the sampling method in that, in the algorithm, details of the processing depends on the "possibility of collision" and no more detailed calcula­tion than that which is necessary is performed.