Trace and Pace Controllable Pedestrian Animation via Guided Trajectory Diffusion (research.nvidia.com)

A groundbreaking method for generating controllable pedestrian animations and trajectories, named TRACE (Trajectories from Controllable Pedestrians), has been introduced. This technique leverages guided diffusion modeling to allow users to define specific goals such as target waypoints, desired speeds, and social group placements, all within a simulated environment. By integrating TRACE with a physics-based humanoid controller called PACER, the system not only achieves natural and realistic animations for a variety of body types but also effectively navigates complex terrains and obstacles. This closed-loop system continuously updates its planning based on simulated results, enhancing the realism of crowd simulations. The significance of this development lies in its potential applications for generating synthetic data, particularly in urban settings, which is crucial for training and testing autonomous vehicles. By facilitating sophisticated trajectory control and realistic animations, this approach may advance the field of robotics and AI by providing a robust framework for simulating human behavior in various environments. The analytical loss functions used in TRACE, along with the terrain-aware capabilities of PACER, further ensure that the generated trajectories are not only accurate but also conducive to collision avoidance and social interaction, making it a versatile tool for future research in AI/ML.