GB2454664A

GB2454664A - Voice Actuated Robot

Info

Publication number: GB2454664A
Application number: GB0722251A
Authority: GB
Inventors: Sandor Mihaly Veres; Aron Gabor Veres
Original assignee: Individual
Current assignee: Individual
Priority date: 2007-11-13
Filing date: 2007-11-13
Publication date: 2009-05-20
Also published as: GB0722251D0

Abstract

A robot or robotic device that operates by computer control in a way that all the details of operations are described by natural language sentences that unambiguously compile into computer code. The natural language sentences are used to give insight to the user on how the robot works. Users can then fully understand sensing, perception, world modelling, goal formulation and decision making of the robot by reading the natural language sentences that corresponds to the computer code of the robot. Users who know the capabilities of this robot from its natural language description can then use it more reliably and can understand its limitations and can improve it or reconfigure it themselves.

Description

DESCRIpTION 2454664

Title: TRANSPARENT ROBOTS AND ROBOTIC DEVICES

1. Setting out the background

(001) The invention falls in the area of robots and robotic devices such as humanoid robots, garden robots, interactive toys, robotic pets, household robots or ground vehicles such as planetary rovers, autonomous underwater vehicles and autonomously controlled boats, spacecrafts, missiles, autonomously controlled aerial vehicles, etc. Whenever "robot" is mentioned in this document it is meant to cover these various applications of robots.

(002) Most existing autonomously controlled robots and vehicles have a command language that can be used to instruct them to do something. These traditional approaches to autonomous robots hide the way the robots operate so their human users are unclear about the processes involved in sensing, action-taking and decision making by the robot.

This uncertainty can make them uncertain to use where reliability and correct application is important. The present invention addresses this problem by inventing the "functionally transparent robot" that has a clear formulation of every detail of its functionality.

(003) There is no patent or any published proposal of this kind at present. There are natural language interfaces to robots but the essence of the described invention is not communication but shared "understanding" between the robot and its users in terms of functionality (004) The presented invention is a new kind of robot that has functional transparency to humans. This principle of transparency could be successfully applied to computer software but the present invention is not claiming application to the area of software, the invention is only concerned with a functionally new physical robot. (005)

1.1 Prior arts

(005) Natural language interfaces to machines described in various inventions such as US 2005/0289134 Al, WO 2005/062202 A3, US 2005/794050A. These connect arbitrary natural language sentences with meaning in some knowledge base and as such are totally different from the present invention.

(006) Traditional ways to achieve reliability of robots is by thorough testing of functionality in various tasks. This invention enhances reliability by sharing more knowledge with the robots user on its operation.

2. Human need that the invention satisfies (007) Even if testing is performed the problem of a communication bottleneck between robots and humans remain: the human user can be uncertain about the ways of behaviour and operations of the robot as simple commands cannot convey information on details of how the command is executed, commands do not reveal possible weaknesses and dangers.

(008) The invention satisfies a human need to operate service robots, autonomous vehicles safely by involving human understanding in every part of their functionality. This way the human can make responsible decisions with regards to a robot's suitability and the ways in which to employ it in various situations.

(009) The invention removes the uncertainty of suitability of the robot in a particular application by the use of complete natural language descriptions of sensing, decision making and action taking procedures of the robot 3. What the invention does (0010) The invention introduces the concept of the "transparent" robot and refers to the robots operational qualities. All the details of sensing, action and decision making operations are described by natural language sentences that prescribe the actual device instructions issued to the robot.

4. Sum mary of the essential features (012) A "transparent robot" is a robot that is functionally organised in a way that every piece of its operation, including sensing, decision making, behaviour rule and action taking are described in natural language sentences that unambiguously correspond or compile into computer code and provide instructions that drives the mechanics of the robot.

(013) The natural language sentence structures are prescribed for the robot and kept in its memory. There is no complex algorithm needed for matching sentence meanings involved, compilation is exact to the meaning.

(014) The usefulness of this approach is not obvious and hence there is no prior invention of this kind but usefulness becomes apparent while using such a system in practice.

5. Drawings used to explain the operation of the components of the invention (0015) Figure 1: Description of the robot architecture 4 1 -CPU(s) that execute the computer program parts corresponding to sentences 4 2 -Central memory (or memories) that contain the programs compiled from sentences and also the compiler itself plus possibly an operating system that can run several processes to execute code compiled -essentially it runs behaviour code by compiling natural language sentences -3 -Digital storage devices. They keep all sentence structures and sentence codes for compilation, keep all behaviour-rule libraries, keep all world knowledge data of the robot in terms of sentences 4 4. -Sensors such as accelerometers, gyros, cameras, microphones, altimeters, inclinometers, magnetometers, or any knid of software that is interfaced to the

CPU

4 5 -Communication ports for sensor and actuator devices and for inter-process communications 4 6 -Network communications devices: radio data links, (W)LAN, Internet, etc. 4 7 -Power unit contains voltage regulators and batteries that can be supplied by renewable energy from the environment (solar, wind, vibration energy harvesting, etc.) or can be charged from electricity distribution networks using suitable charger unit.

6. Illustrations of the invention The following examples illustrate the functional descriptions that transparent robots may have and where each sentence unambiguously compiles into computer code.

6.1 A household robot These are some parts of descriptions of functionality in natural language sentences that unambiguously convert into a computer program controlling robot behaviour: Continuously use cameras A, B, C to produce 3D scene model M. Meet people present from M: Identify people set P from scene model M: Use fast masking' and Gabor wavelets to identify people set P from A, B and C. Use people memory PM to recognise people in set P: Use matching procO89l' to identify known people K in P from PM.

Identify people KN in P whom you do not know using K and PM.

Extract features F of people in NK in P whom you do not know.

Add people NK and features F to your people memory PM.

Ask the names Nms of people in NK. Add Nms to memory PM.

Recognise physical environment from M: Identify object set 0 from scene model M. Model physical contacts Pc between the objects in set 0.

Identify objects MO you can move in 0 using Pc and object weight estimates in 0 and your hand grasping capabilities.

Identify set of path options Pths you may be able to take in scene model M. Collect your action options in your capability set Caps.

Listen to commands with your name: Get data stream AuD from microphones Mid and Mic2.

Continuously analyse AuD for human voice: Use procedure markovO9l' to filter out speech segments Sps.

Look for your name "Robi" in Sps to get command C. If command C is nonempty then ask human whether C is that you are supposed to do.

Wait for answer A from human: Look for your name "Robi" in Sps to get answer A. If answer A is nonempty then decide if answer R was yes'.

If R was yes then carry out command C, otherwise wait for repeat of command C: Look for your name "Robi" in Sps to get command Cs.

If command Cs is nonempty then ask human whether Cs is you are supposed to do.

If answer A is nonempty then decide if answer R was yes'.

If answer R was no the do nothing and continue.

If your command C is non-empty then execute command C. Execute command C. Interpret command C in terms of modified model Mgoal as obtained from M. Use your cababilities Caps to derive plan P2goaI to create Mgoal while strictly confining to behaviour Rule_book_of_robots.

Carry out plan P2goaI and do not let listen to new commands until you finished.

Rules contained in Rule_book_of_robots: Do not cause any objects to fall to the ground.

Do not touch any human being in any way.

If touched by a human then stay still until human moves away. Etc.

The above example description is not a manual or description for the user. It compiles into computer code controlling the robot. The user of the robot can (1) Insert sentences (2) modify sentences (3) replace sentences by ones provide in a Sentence Library for the robot provided by its manufacturer or (4) erase or reorder sentences of robot behaviour.

7.2 Autonomous aerial vehicle (AAV) When started up stabilise engine running within 5 minutes.

Continuously receive radio commands into S. Identify if S is of type tS as question', command' or statement'.

If tS is question' then do the following: Match tS to your Questions_library to get question key qK.

Use qK and model M to prepare answer A. Announce answer A suitably to the situation.

If tS is command' then do the do the following: State your interpretation of tS and ask confirmation.

If tS is about mission' then get detailed plan in place of tS.

Produce action plan PIn for tS under constraints from "RuIe_book_of_UAVs'.

Carry out plan Pin and do not listen to new commands until you finished.

If tS is statement' then do the following:

Say to owner Ow the word "really? OK, I got that!'.

Interpret tS and update scene model M. Update memory Mall with parts from scene model.

State your interpretation of tS and ask confirmation.

Match command tS with meaning Mts in file "UAV_commands_library.lib'.

Get detailed plan in place of tS: Ask human operator for detailed written plan P1.

Interpret P1 to get your LongPth.

Rule book of AUVs includes: While on mission' keep to the following: Monitor your tracking of your LongPth and inform mission control if you are off track' by -lOOm.

Continuously monitor potential obstacle set Obs.

Replan your immediate path Pth every -5s to follow missing track and avoid Obs.

Monitor mission task opportunities. Etc

Each of these sentences compiles into a piece of computer code that the robot uses. Any of these can be replaced by sentences from the robots sentence library as provided by its manufacturer.

The user of the AAV can read this natural language text and hence gains insight into how it works, no instruction manual needed to understand the operation of the AAV.

Claims

Any kind of robot or robotic device (including memory, sensors, electromechanical actuators and interface or perception hardware components) that operates by interpreting unambiguous natural language sentences into instructions (a program) that govern its planning, behaviour rules and functionality as defined by the afore mentioned natural language sentences, which completely describe the robots operation and are perfectly understandable to humans and completely understandable to the robot device through a chain of compilations and interpretations until a device level instructions are reached; resulting in a robot device reliably programmable in a simple way through prescribed sentences that safely allow for changes in operation and allowing anyone to tell the robot how to behave via simple natural language.