Presentation is loading. Please wait.

Presentation is loading. Please wait.

Using Expectations to Drive Cognitive Behavior Unmesh Kurup Christian Lebiere, Tony Stentz, Martial Hebert Carnegie Mellon University.

Similar presentations


Presentation on theme: "Using Expectations to Drive Cognitive Behavior Unmesh Kurup Christian Lebiere, Tony Stentz, Martial Hebert Carnegie Mellon University."— Presentation transcript:

1 Using Expectations to Drive Cognitive Behavior Unmesh Kurup Christian Lebiere, Tony Stentz, Martial Hebert Carnegie Mellon University

2 Cognitive Decision Cycle t+1 Calculate Mismatch World High-level Cognition Retrieve Response World Action Prediction t-1t Cognition Cognition is driven by Expectations/Predictions.

3 Pedestrian Tracking & Behavior Classification Goals: Investigate use of expectations Integrate with perception Run both offline & real-time

4 Integrated System

5 Partial Matching & Blending Chunk2 isa location- chunk id person2 nextx 1010 nexty 500 Chunk3 isa location- chunk id person3 nextx 187 nexty 313 Chunk4 isa location- chunk id person1 nextx 299 nexty 100 +retrieval> isa location- chunk id person1 nextx 300 Chunk1 isa location- chunk id person1 nextx 255 nexty 100 Chunk4 isa location- chunk id person1 nextx 299 nexty 100 Declarative Memory Partial Matches Chunk5 isa location- chunk id person1 nextx nexty 100 Blended result Chunk1 isa location- chunk id person2 nextx 255 nexty 100

6 Using Expectations: Tracking Chunk-type visual-location idX YDxDyNextxNexty Foreach Object o: +blending> isa visual-location id o compare to (x,y)s from perception pick thresholded closest match, calculate newdx, newdy, newx, newy +imaginal> isa visual-location id o …

7 Features straight1 straight2detourleftstraight3veer Features: BehaviorFeatures Normal – Straightstraight1, straight2, straight3 Normal – Leftstraight1, straight2, left Peekstraight1, detour, left, no-chk-pt BehaviorFeatures Detourstraight1, detour, straight3, chk-pt Veerstraight1, straight2, left, veer, chk-pt Walkbackstraight1, straight2, left, straight2, straight1, chk-pt

8 Using Expectations: Detecting Features from Data Straight & Left Deviation from expected location indicates a point of interest

9 Foreach location +blending> isa visual-location x =x y =y compare to (x,y)s from perception if path deviates more than threshold, mismatch! +imaginal> isa visual-location id o … Cluster points into regions

10 Detected Features

11 Data Combined Arms Collective Training Facility(CACTF) at Fort Indiantown Gap, PA. 4 examples. 3/1 split. Multiple behavior set – 10 behaviors.

12 Behaviors Straight & Left Peek Detour Veer Walkback

13 Results Hand-coded Model (Single Behavior Set) Hand-coded Model (Multiple Behavior Set) Made99.3%Made46.5% Correct99.15%Correct30.2% Incorrect0.15%Incorrect16.3% Learning Model (Single Behavior Set) Learning Model (Multiple Behavior Set) Made86.1%Made82.4% Correct68%Correct43.8% Incorrect18.1%Incorrect38.6%

14 Future Work – Semantic Labels

15 Future Work – Using Semantic Labels BehaviorFeatures (Spatial)Features (Semantic) Normal – Straightstraight1, straight2, straight3 Sidewalk, Pavement Normal – Leftstraight1, straight2, left Sidewalk Peekstraight1, detour, left, no-chk-pt Pavement, Sidewalk Detourstraight1, detour, straight3, chk-pt Pavement Veerstraight1, straight2, left, veer, chk-pt Sidewalk, Pavement Walkbackstraight1, straight2, left, straight2, straight1, chk- pt Sidewalk

16 Future Work Generic model of monitoring using expectations Learn expectations Monitor for deviations from expectations – Signal failure – Provide for recovery

17 Collaborators Max Bajracharya, JPL Bob Dean, GDRS Brad Stuart, GDRS FMS lab, CMU


Download ppt "Using Expectations to Drive Cognitive Behavior Unmesh Kurup Christian Lebiere, Tony Stentz, Martial Hebert Carnegie Mellon University."

Similar presentations


Ads by Google