Stanford Robotics Seminar ENGR319 | Winter 2026 | Autonomous Navigation in Outdoor Environments

Robot-Specific Traversability Analysis

Navigation requirements vary significantly by platform, as wheeled robots require flat surfaces while legged robots can traverse curbs, stairs, and vegetation depending on their size and weight.

Gaussian Splatting with Physical Properties

The system renders semantic environments using Gaussian splatting fused with LiDAR Euclidean Signed Distance Fields, jointly estimating material types and physical properties including friction, hardness, stiffness, and density.

VLM-Based Trajectory Selection

Multiple trajectory candidates are generated via autoencoder-decoder mechanisms with Gaussian projections, then ranked and selected by Vision Language Models based on goal proximity, outperforming purely learning-based approaches like Nomad.

Three-Stage Social Navigation Framework

The task is decomposed into perception (identifying pedestrians/groups), prediction (forecasting motion), and action (planning compliant movements) to ensure pedestrian comfort and traffic rule adherence.

Fine-Tuned Vision Language Models

Fine-tuned LLaVA models outperform ChatGPT and Gemini in social navigation benchmarks, particularly in predicting pedestrian behaviors and avoiding actions like intersecting groups or racing against people.

Traffic Context Understanding

Robots must interpret traffic lights, right-of-way rules, and vehicle interactions to navigate urban environments compliantly rather than simply avoiding obstacles.

Demographic Imperative for Elder Care

With 55 million Americans aged 65+ in 2020 and projections of 20% senior population by 2030, companion robots aim to enhance independence through mobility assistance and health monitoring.

Adaptive Pace-Moving Navigation

Unlike traditional social navigation that avoids humans, companion robots must adaptively match walking pace with older adults, process language instructions, and provide verbal alerts about hazardous terrains.

Behavior Analysis and Fall Prevention

Robots will monitor movement patterns to correlate motions with disease risks and predict falls while supporting independent exercise, addressing seniors' desire to avoid traditional walkers.