Robot Execution Failures

Data Type

multivariate time series

Abstract

This dataset contains force and torque measurements on a robot after failure detection. Each failure is characterized by 15 force/torque samples collected at regular time intervals starting immediately after failure detection.

Sources

Original Owner and Donor

Luis Seabra Lopes and Luis M. Camarinha-Matos
Universidade Nova de Lisboa, 
Monte da Caparica, Portugal
Date Donated: April 23, 1999

Data Characteristics

The donation includes 5 datasets, each of them defining a different learning problem:

Feature information

All features are numeric although they are integer valued only. Each feature represents a force or a torque measured after failure detection; each failure instance is characterized in terms of 15 force/torque samples collected at regular time intervals starting immediately after failure detection; The total observation window for each failure instance was of 315 ms.

Each example is described as follows:
                 class
                 Fx1	Fy1	Fz1	Tx1	Ty1	Tz1
                 Fx2	Fy2	Fz2	Tx2	Ty2	Tz2
                 ......
                 Fx15	Fy15	Fz15	Tx15	Ty15	Tz15

where Fx1 ... Fx15 is the evolution of force Fx in the observation window, the same for Fy, Fz and the torques; there is a total of 90 features.

Number of instances in each dataset

   -- LP1: 88
   -- LP2: 47
   -- LP3: 47
   -- LP4: 117
   -- LP5: 164

Class Distribution

   -- LP1: 24% normal
           19% collision    
           18% front collision
           39% obstruction
   -- LP2: 43% normal
           13% front collision
           15% back collision
           11% collision to the right
           19% collision to the left
   -- LP3: 43% ok
           19% slightly moved
           32% moved
            6% lost
   -- LP4: 21% normal
           62% collision
           18% obstruction
   -- LP5: 27% normal
           16% bottom collision
           13% bottom obstruction
           29% collision in part
           16% collision in tool

Other Relevant Information

Feature transformation strategies

In order to improve classification accuracy, a set of five feature transformation strategies (based on statistical summary features, discrete Fourier transform, etc.) was defined and evaluated. This enabled an average improvement of 20% in accuracy. The most accessible reference is [Seabra Lopes and Camarinha-Matos, 1998].

Data Format

The data is stored in five ASCII files, with each example in the files described as follows:
                 class
                 Fx1    Fy1 Fz1 Tx1 Ty1 Tz1
                 Fx2    Fy2 Fz2 Tx2 Ty2 Tz2
                 ......
                 Fx15   Fy15    Fz15    Tx15    Ty15    Tz15
where Fx1 ... Fx15 is the evolution of force Fx in the observation window, the same for Fy, Fz and the torques; there is a total of 90 features.

Past Usage

Seabra Lopes, L. (1997) "Robot Learning at the Task Level: a Study in the Assembly Domain", Ph.D. thesis, Universidade Nova de Lisboa, Portugal.

Seabra Lopes, L. and L.M. Camarinha-Matos (1998) Feature Transformation Strategies for a Robot Learning Problem, "Feature Extraction, Construction and Selection. A Data Mining Perspective", H. Liu and H. Motoda (edrs.), Kluwer Academic Publishers.

Camarinha-Matos, L.M., L. Seabra Lopes, and J. Barata (1996) Integration and Learning in Supervision of Flexible Assembly Systems, "IEEE Transactions on Robotics and Automation", 12 (2), 202-219.


The UCI KDD Archive
Information and Computer Science
University of California, Irvine
Irvine, CA 92697-3425
Last modified: March 11, 1999