saltylab-firmware/jetson/ros2_ws/src/saltybot_social_memory_node

SaltyBot Social Memory Database (Issue #443)

Persistent SQLite-backed social knowledge system for maintaining person profiles, relationship history, and interaction statistics.

Features

1. Person Profile Management

• Person Records: Name, relationship tier, custom greetings, notes, face embeddings
• Relationship Tiers:
  • Stranger: 0 encounters (default), cold greetings
  • Regular: 5+ encounters, familiar greetings
  • Favorite: 20+ encounters (or 80%+ positive interactions), warm greetings
• Custom Greetings: Tier-specific greeting strings
• Face Embeddings: Optional vector storage for face recognition
• Notes: Free-form observations about the person

2. Encounter Tracking

• Encounter Recording:
  • Timestamp, transcript of conversation, detected mood
  • Engagement score (0.0–1.0)
  • Positive/negative interaction flag
• Automatic Tier Promotion:
  • Encounter count-based: 5→regular, 20→favorite
  • Quality-based: 80%+ positive interactions required
  • Cooldown: 1 hour minimum between promotions
• Statistics Aggregation:
  • Total encounter count per person
  • Positive interaction ratio
  • Relationship score (computed from interactions)

3. ROS2 Service Interfaces

/saltybot/social_memory/lookup

Request:
  query_type: "id" | "name" | "recent"
  query_value: person_id | name_pattern | days_back

Response:
  found: boolean
  person_id: int32
  name: string
  relationship_tier: string
  relationship_score: float32 (0.0–1.0)
  interaction_count: int32
  custom_greetings: string[3]  # [stranger, regular, favorite]
  notes: string
  last_seen: Time

/saltybot/social_memory/update

Request:
  person_id: int32
  name: string
  relationship_tier: string
  custom_greetings: string[3]
  notes: string
  embedding: float32[]

Response:
  success: boolean
  message: string
  updated_person_id: int32

/saltybot/social_memory/encounter

Request:
  person_id: int32
  transcript: string
  mood: string
  engagement_score: float32
  positive_interaction: boolean

Response:
  success: boolean
  message: string
  encounter_id: int32
  tier_changed_to: string (or empty if no change)

/saltybot/social_memory/stats

Request:
  (empty)

Response:
  total_persons: int32
  total_encounters: int32
  favorites_count: int32
  regulars_count: int32
  strangers_count: int32
  database_size_mb: float32
  last_backup: Time
  database_path: string

4. Statistics Publishing

• Topic: /saltybot/social_memory/stats_update (std_msgs/String, JSON)
• Frequency: Configurable (default 60s)
• Content: Database statistics snapshot

5. Backup & Restore

• Automatic Backups: Gzipped snapshots stored in backup directory
• Backup Tracking: Recorded in database with timestamp and size
• Manual Restore: Restore from any backup file

Database Schema

persons Table

Column	Type	Notes
person_id	INTEGER PRIMARY KEY	Auto-increment
name	TEXT UNIQUE	Person's name
embedding	BLOB	JSON-encoded face embedding
relationship_tier	TEXT	stranger/regular/favorite
relationship_score	REAL	0.0–1.0 computed from encounters
interaction_count	INTEGER	Total encounters
positive_interactions	INTEGER	Count of positive encounters
greeting_stranger	TEXT	Greeting for strangers
greeting_regular	TEXT	Greeting for regular contacts
greeting_favorite	TEXT	Greeting for favorites
notes	TEXT	Custom observations
created_at	TIMESTAMP	When person was first recorded
last_seen	TIMESTAMP	Last encounter timestamp
last_tier_promotion	TIMESTAMP	Cooldown tracking

encounters Table

Column	Type	Notes
encounter_id	INTEGER PRIMARY KEY	Auto-increment
person_id	INTEGER	Foreign key to persons
transcript	TEXT	Conversation text/summary
detected_mood	TEXT	Emotion detected during encounter
engagement_score	REAL	0.0–1.0
positive_interaction	BOOLEAN	Was interaction positive?
encounter_timestamp	TIMESTAMP	When encounter occurred

backups Table

Column	Type	Notes
backup_id	INTEGER PRIMARY KEY	Auto-increment
backup_path	TEXT UNIQUE	Full path to backup file
backup_timestamp	TIMESTAMP	When backup was created
file_size_mb	REAL	Compressed size in MB

Configuration

Edit config/social_memory.yaml:

database_path: "/home/seb/saltybot-data/social_memory.db"
backup_dir: "/home/seb/saltybot-data/backups"

publish_stats: true
stats_publish_interval: 60.0  # seconds

tier_promotion:
  regular_threshold: 5        # Encounters for regular tier
  favorite_threshold: 20      # Encounters for favorite tier
  positive_ratio_min: 0.8     # 80% positive required
  cooldown_hours: 1           # Minimum between promotions

Running

Launch the service node

ros2 launch saltybot_social_memory_node social_memory.launch.py

Direct execution

ros2 run saltybot_social_memory_node social_memory

Usage Examples

Add a new person

ros2 service call /saltybot/social_memory/update saltybot_social_memory/PersonUpdate \
  "{name: 'Alice', relationship_tier: 'stranger', \
    custom_greetings: ['Hello there!', 'Nice to see you!', 'Great to see you again!']}"

Record an encounter

ros2 service call /saltybot/social_memory/encounter saltybot_social_memory/EncounterRecord \
  "{person_id: 1, transcript: 'Alice: Hi! How are you?', \
    mood: 'happy', engagement_score: 0.85, positive_interaction: true}"

Lookup a person

ros2 service call /saltybot/social_memory/lookup saltybot_social_memory/PersonLookup \
  "{query_type: 'name', query_value: 'Alice'}"

Get database statistics

ros2 service call /saltybot/social_memory/stats saltybot_social_memory/DatabaseStats

Monitor statistics stream

ros2 topic echo /saltybot/social_memory/stats_update

Integration Points

1. Emotion Engine (saltybot_emotion_engine)

   • Queries person familiarity for warmth modifier
   • Records encounters to build relationship history

2. Person Tracker (sl-perception)

   • Provides person_id from face recognition
   • Supplies engagement_score and detected_mood

3. Voice Recognition (sl-jetson)

   • Provides transcript of conversations
   • Records as encounter transcript

4. Greeting System

   • Queries custom_greetings based on tier
   • Personalizes robot responses

Tier Promotion Logic

When encounter recorded:
  1. Increment interaction_count
  2. Update positive_interactions if positive_interaction=true
  3. Compute positive_ratio = positive_interactions / interaction_count
  4. Check promotion eligibility:
     - current_tier == "stranger" && interaction_count >= 5 && positive_ratio >= 0.8
       → Promote to "regular"
     - current_tier == "regular" && interaction_count >= 20 && positive_ratio >= 0.8
       → Promote to "favorite"
  5. If promoted and cooldown_hours passed since last promotion:
     - Update relationship_tier
     - Set last_tier_promotion = now()
     - Return tier_changed_to in response

Backup Strategy

• Manual backups via backup_database() method
• Automatic tracking in backups table with timestamp
• Restore from any backup: restore_database(backup_path)
• Format: gzip-compressed SQLite database (.db.gz)

Thread Safety

• All database operations protected by threading.Lock()
• Safe for concurrent ROS2 service calls
• Sequential writes to ensure data integrity

Performance Characteristics

• Lookup: O(1) indexed by person_id, O(n) by name pattern
• Encounter recording: O(1) insertion + O(1) stats update
• Database size: ~1KB per person + ~0.5KB per encounter

Future Enhancements

1. Persistent memory integration with LLM (e.g., "Alice loves plants")
2. Group relationships (track person groups/families)
3. Preference learning (dietary restrictions, favorite topics)
4. Sentiment analysis over time
5. Web UI for person management
6. Integration with voice recognition confidence
7. Automated cleanup of old encounters (>1 year)
8. Graph-based relationship networks