Code & Data for "Fatigue-Aware Sub-Second Combinatorial Auctions for Dynamic Cycle Allocation in Human–Robot Collaborative Assembly". This paper introduces an adaptive combinatorial auction framework for dynamic task allocation in human-robot col

This paper presents an adaptive combinatorial auction-based dynamic cycle allocation framework quantified in active auction cycles·min⁻¹ for Human–Robot Collaboration in industrial assembly, designed for sub-second responsiveness to fluctuating human fatigue and robot availability. A human operator and a FANUC LR Mate 200iD-7L robot submit real-time bids integrating cycle execution time, predicted energy consumption, and instantaneous human fatigue—estimated from wearable-inspired IMU and heart-rate synthetic sensor streams—to reflect ergonomic state and task complexity. The allocation objective is formulated as a binary-decision optimization problem, maximizing total system utility under the constraint that each 100 ms auction cycle is assigned to at most one agent. To solve this NP-hard problem in real time, a greedy approximation algorithm with a 0.632 guarantee and <i>O</i> per-cycle complexity is employed, operating at a 10 Hz update rate to ensure continuous adaptation to state changes. High-fidelity RoboDK v5.9 simulations demonstrate robust performance: a mean active-cycle throughput of 360 ± 5 cycles·min⁻¹ (≈ 60 % robot engagement), an average fatigue of 0.50 ± 0.01 units per cycle (4 % improvement), and a collision-free rate of 99.85 ± 0.05 %, compared to a static rule-based baseline. Statistical analysis via one-way ANOVA and Tukey HSD (α = 0.05) confirms significance (p &lt; 0.05) with large effect sizes. This work delivers a transparent, scalable mechanism for Industry 5.0 applications requiring human-centered, real-time automation. Future research will explore physical testbeds, multi-agent extensions, and integration of sustainability KPIs such as energy-per-cycle and CO₂ footprint estimation.