Market-Driven Multi-Operators Coordinated effort in Robot Soccer Space.

Uploaded on:
Category: Home / Real Estate
Robot soccer is a very much characterized environment which gives a decent proving ground to ... Every robot has basic, obviously characterized activities accessible and the general assignment ...
Slide 1

Market-Driven Multi-Agent Collaboration in Robot Soccer Domain

Slide 2

Today\'s Presentation Multi-Agent Systems Robot Soccer The Market Methodology Market-Driven Approach Reinforcement-Based Market-Driven Approach "Another Approach"

Slide 3

Multi-Agent Systems

Slide 4

Multi-Agent Systems Why use multi-operators frameworks? Multi-specialists frameworks are turning out to be more famous than complex single operators frameworks since they dispose of the issue of single purpose of disappointment.

Slide 5

Multi-Agent Systems How would they function? Multi-operators frameworks work by breaking down an unpredictable undertaking into a few low-level activities which can then be doled out to the individual colleagues.

Slide 6

Multi-Agent Systems How to allocate assignments? This is a key issue, the framework must separate the errands and direction the group such that the group aggregately finishes the general undertaking.

Slide 7

Multi-Agent Systems How to allocate undertakings? The framework must monitor every robot\'s abilities (inconsequential in a homogeneous group, however more convoluted in a heterogeneous group)

Slide 8

Robot Soccer

Slide 9

Robot Soccer Problem Domain? We will take a gander at robot soccer as the issue area as it gives a decent true space for creating multi-operators frameworks.

Slide 10

Robot Soccer Domain Robot soccer is a very much characterized environment which gives a decent proving ground to creating multi-operators systems. Every robot has straightforward, obviously characterized activities accessible and the general errand straightforward –Beat the other group.

Slide 11

Robot Soccer Domain Robot soccer gives a decent method for looking at two frameworks/methodologies. The two frameworks can essentially be played against each other and see which group wins the most matches.

Slide 12

The Problem We require a method for organizing the robots to each perform an errand/satisfy a part (ie assault, support, shield, goalie and so forth). The Market-Driven Approach for planning the multi-specialists framework depends in transit free-advertises amplify benefits.

Slide 13

The Market Methodology

Slide 14

The Market Methodology The fundamental objective in free-markets is the amplification of the general benefit. The hypothesis is that if every member in the business sector tries to augment its benefit, the general benefit ought to increment.

Slide 15

Market-Driven Approach

Slide 16

The Market-Driven Approach The Market-Driven Approach parts up the principle errand into straightforward undertakings and a bartering is then held for every assignment. The robots work out the expense for them to perform an assignment and after that put in their best offer to the salesperson. The robot which puts in the most minimal offer gets the task.

Slide 17

The Market-Driven Approach

Slide 18

The Market-Driven Approach In Robot Soccer a closeout is held for each of the distinctive parts. The robots compute the expense of satisfying those parts (in light of separation to ball and so forth) and offer on them. The robots with the best offer on every part will be doled out the part.

Slide 19

The Market-Driven Approach Two (or more) robots may get the same task where they should coordinate to perform the assignment (ie a robot with the ball assaults the objective and another robot bolsters it by driving not far behind)

Slide 20

The Market-Driven Approach leverage of the Market-Driven Approach is that every robot figures the expense of performing every part and imparts that cost to alternate robots. This cost worth is much less demanding and snappier to impart as opposed to sending the majority of the measurements to alternate robots.

Slide 21

The Market-Driven Approach What about how the bartering is run? Unified Distributed Hybrid

Slide 22

Centralized There exists an expert specialists (barker) that controls the barterings and doles out the parts. The expert specialists gets offers from every other operators for every assignment and sends the closeout results back. Computationally proficient. Inclined to single point disappointments.

Slide 23

Distributed No expert specialists. Each specialists telecasts its offer for each assignment. Each specialists runs the same closeout component and parallely figures the sale results. Powerful against single point disappointments Requires more calculation altogether.

Slide 24

Hybrid There exists an expert operators There is additionally a sale for the assignment of being the expert Robust against single point disappointments Computationly effective Still not executed, no test outcomes.

Slide 25

The Market-Driven Approach Problem – How to compute the expenses? Every robot must have the capacity to figure the expense of filling a specific part. The settings for the cost counts must be adjusted, the execution of the framework relies on upon the alignments being right. Eg. - C aggressor = M 2 *dist Ball + M 2 *dist OppGoal

Slide 26

Reinforcement-Based Market-Driven Approach

Slide 27

Reinforcement Learning Reinforcement-Based Market-Driven Approach makes utilization of Reinforcement Learning (RL) to take in the part task process. RL is utilized when the specialists is educated about the results of its activities. RL replaces the part task as depicted previously.

Slide 28

Reinforcement Learning With the RL framework, the robot nearest to the ball doles out itself as the assailant and the remaining operators (barring the static goalie) allocate themselves as indicated by a state vector. (see next slide)

Slide 29

Reinforcement Learning The Rules: Goalie is statically allocated. The Robot nearest to the ball is doled out the part of aggressor. Alternate robots are relegated parts by a state vector. State vector measurements – separations to the ball, objectives, robots, the cost values and the nearest player to the ball.

Slide 30

Reinforcement Learning Broadcast Position and Cost Data Calculate Attack Cost Array Calculate Defense Cost Array Closest to Ball Cheapest Yes Shoot No Role Assigned According to Cost Value Pass To Cheapest

Slide 31

New Approach

Slide 32

New Approach The New Approach is viably a streamlined adaptation of the Reinforcement Learning framework. However as opposed to utilizing the definite positions of the robots, the field is partitioned into a lattice.

Slide 33

New Approach

Slide 34

New Approach The framework can now utilize this network to settle on a choice on what part the robot ought to be performing. To appoint parts, the framework utilizes a state vector with the accompanying measurements: Ball Position (network number), Ball Possession, Current Role relegated by the Market-Driven methodology, Teammate positions and Opponent positions.

Slide 35

New Approach This methodology joins the Market-Driven Approach and the Reinforcement Learning based group with the network partition of the board to keep the quantity of variables in the state vector to a base.

Slide 36


Slide 37

Results The New Approach which joins the Market-Driven, RL and matrix framework out performs the greater part of alternate groups reliably more than 90 matches.

Slide 38


Slide 39

References Kose, H., Kaplan, K., Mericli, C., Tatlidede, U. & Akin, L. (2005). Market-Driven Multi-Agent Collaboration in Robot Soccer Domain. Bleeding edge Robotics, 407-416. Kurt , B. (2007). Bogazici University Robotics Server. Recovered September 09, 2007, from

View more...