ReTracing: An Archaeological Approach Through Body, Machine, and Generative Systems

2025-08-09 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Computer Vision · Depth: Advanced, long

Summary

ReTracing is a multi-agent embodied performance art project that employs an archaeological approach to investigate how AI influences and produces bodily movement. The project extracts sentences describing human-machine interaction from seven science-fiction novels, including "Frankenstein" and "Klara and the Sun." Large language models, specifically Qwen-2.5 with a temperature of 0.7, generate paired "what to do" and "what not to do" prompts from these excerpts. A diffusion-based text-to-video model, Wan Team (2025), converts these prompts into choreographic guides for a human performer and motor commands for a quadruped robot. Both agents perform on a mirrored floor, with their actions captured by multi-camera motion tracking and reconstructed into 3D point clouds and motion trails, forming a digital archive. This process aims to reveal how generative systems encode socio-cultural biases through choreographed movements and explores the meaning of humanity alongside intelligent, moving AI.

Key takeaway

For AI scientists exploring the societal impact of generative models, ReTracing demonstrates a novel method to visualize algorithmic bias. Your research can adopt this "Archaeology of AI" framework to deconstruct how LLMs and diffusion models interpret and reshape human actions, particularly concerning embodied identity. Consider using multi-agent performance and motion tracking to make abstract biases concrete, providing a compelling visual critique of AI's internal logic and its ethical implications.

Key insights

ReTracing uses AI to choreograph human and robot movements, revealing how generative systems encode socio-cultural biases.

Principles

• AI can act as an archaeological tool.
• Generative systems inscribe control through prompts.
• Movement can reveal algorithmic logic.

Method

Literary excerpts are processed by LLMs to create positive and negative movement prompts. These prompts guide a text-to-video model for human choreography and direct robotic actions, with movements then tracked and archived as 3D data.

In practice

• Use LLMs for creative prompt generation.
• Combine text-to-video with robotics for embodied art.
• Track 3D motion for digital performance archives.

Topics

• Generative AI
• Human-Robot Interaction
• Algorithmic Bias
• 3D Motion Tracking
• Embodied Performance

Best for: AI Scientist, Research Scientist, AI Researcher, Robotics Engineer, AI Ethicist

Related on AIssential

• Embodied-BenchClaw: An Autonomous Multi-Agent System for Embodied Spatial Intelligence Benchmark Construction — Embodied-BenchClaw automates the creation and maintenance of embodied spatial intelligence benchmarks using an agentic system.
• A Mechanistic Analysis of Sim-and-Real Co-Training in Generative Robot Policies — Co-training effectiveness in robotics stems from structured representation alignment and importance reweighting.
• Beyond Waypoints: A Trajectory-Centric Waypointing Paradigm for Vision-Language Navigation — Grounding waypoints in executable trajectories resolves planning-control inconsistencies and improves reachability in Vision-Language Navigation.
• GEM: Generative Supervision Helps Embodied Intelligence — Integrating depth map generation into VLM pre-training significantly enhances embodied intelligence by bridging semantic and physical knowledge gaps.
• X-Diffusion: Training Diffusion Policies on Cross-Embodiment Human Demonstrations — X-Diffusion selectively integrates noisy human demonstrations into robot policy training to overcome embodiment mismatches.
• Lifting Embodied World Models for Planning and Control — Lifting world models with low-dimensional waypoints significantly improves embodied agent planning efficiency and accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.