Social Navigation

Scene-aware social robot navigation in corridors and doorways using SocialVAE prediction, LLM trajectory selection, and MPC-style planning.

Social Navigation in Corridor, Doorway, and Slow Corridor Scenes

Project Problem Statement

This project studies how a robot can navigate shared human spaces while respecting scene structure, personal space, and changing social constraints.
The objective is not only to reach a destination, but to do so in a way that is safe, socially acceptable, and context-aware.

The navigation problem is framed as social decision-making: the robot interprets likely human motion, accounts for static boundaries and obstacles, and chooses actions that remain robust across different environments.

Methodology

The system combines three layers:

  1. Scene Simulation
    • Structured environments are created with walls, obstacles, doorway constraints, and different human movement patterns.
    • Each scene is designed to test a specific social navigation challenge.
  2. Human Future Prediction and Social Reasoning
    • Multiple possible future trajectories are generated for each human.
    • A reasoning module evaluates candidate futures based on motion consistency, scene context, and social plausibility.
    • The selected futures are used to guide safe robot behavior.
  3. Robot Planning and Control
    • The robot plans actions using socially aware control logic.
    • Planning considers predicted human motion, static geometry, and bottleneck behaviors (e.g., doorways).
    • Motion continuity, spacing, and conflict avoidance are prioritized.

Scene 1 — Corridor

A narrow passage with limited maneuvering room where humans and the robot share the same space.

Fig 1: Corridor scene setup used for evaluating socially aware navigation in constrained geometry.

What this scene evaluates

  • Safe passing in constrained geometry
  • Avoiding deadlocks and oscillation
  • Coordinating with bidirectional pedestrian flow

Scene 2 — Doorway

A threshold crossing scenario with temporary bottleneck behavior and opposing flows.

Fig 2: Doorway scene setup for timing, yielding, and crossing decisions.

What this scene evaluates

  • Timing at bottlenecks
  • Crossing conflict handling
  • Socially compliant yielding behavior

Scene 3 — Slow Corridor

A tighter corridor where the robot follows a slower human and adapts pace/overtake behavior.

Fig 3: Slow corridor setup for pacing and following-distance behavior.

What this scene evaluates

  • Following distance and pace adaptation
  • Socially acceptable overtaking decisions
  • Comfort-aware motion in narrow shared space

Summary

Across corridor, doorway, and slow-corridor settings, the project demonstrates a socially grounded navigation pipeline that combines environment modeling, multi-future prediction, social reasoning, and constrained planning.

The results show that scene-aware social navigation can improve both safety and interaction quality in shared human environments.