Coordinated Systems Lab

Safe navigation of autonomous agents in human centric environments (or in a multi-agent context in general) requires the ability to understand and predict motion of neighboring agents. However, predicting pedestrian or other agent intent is a complex problem. Motion is governed by complex social navigation norms, is dependent on neighbors’ trajectories, and is multimodal in nature.

Models

• In this work, we propose SCAN, a Spatial Context Attentive Network
• SCAN can jointly predict socially-acceptable multiple future trajectories for all agents in a scene.
• SCAN encodes the influence of spatially close neighbors using a novel spatial attention mechanism in a manner that relies on fewer assumptions, is parameter efficient, and is more interpretable compared to state-of-the-art spatial attention approaches.

Results

• Through experiments on several datasets we demonstrate that our approach can also quantitatively outperform state of the art trajectory prediction methods in terms of accuracy of predicted intent.

This project combines data- and model-driven predictive controllers to solve problems that deal with unknown functions operating interdependently with functions that are explicitly modeled with mathematical functions. In other words, we are combining black-box machine learning techniques with analytical approaches based on 1st-principle physical models.

• our algorithms can cope with very little data and build solutions based on ‘‘learning’’ from the recently generated trajectories
• in other words, it uses recursive approaches that are guaranteed to improve performance in each iteration until converging to an optimal trajectory, typically needing only a few iterations.

This project seeks to answer two questions. First, where does the machine fit into the team in safety-critical operations? Second, how do we assure safety in human-machine teams with increasingly-autonomous capabilities? To answer these questions, the project involves:

• modeling and computer-based fast-time simulation of aviation operational concepts of operation with a wide range of different human-machine team structures. These models identify the actions required of each teammate, the information they depend upon and that they create, and their predicted timing.
• Safety analysis of team operations, including hazard analysis of predictable contingencies requiring a response from the flightdeck team, and then extrapolation of safety parameters from the modeling results given earlier.
• Human-in-the-loop (HITL) experiment results, providing measures similar to those provided by the modeling and computer-based fast-time simulation as observed in the operations of an actual human-machine team (human pilot and autonomy) in actual operation