Auction-Based Scheduling

Agents place all-or-nothing bids on bundles of heterogeneous resources rather than individual time slots. This captures synergistic value—a job might only be profitable if it secures both a CNC machine and a specific curing oven simultaneously.

• Prevents the exposure problem where an agent wins one resource but fails to secure the complementary resource
• Reduces total cost by allowing agents to express complex preferences
• Commonly solved using the Winner Determination Problem (WDP) , an NP-hard optimization

A sealed-bid mechanism where the winning agent pays the marginal harm their win imposes on other agents, not their own bid. This incentivizes truthful bidding as the dominant strategy.

• Eliminates strategic bid shading and gaming
• The payment equals the difference between the optimal social welfare with and without the winning agent
• Critical for supply chain coordination where honest capacity revelation prevents the bullwhip effect

Each agent encodes its scheduling priority into a utility function that maps completion time to economic value. High-priority orders generate steeper discount curves for tardiness.

• Linear decay: Value drops at a constant rate per minute of delay
• Step functions: Value collapses after a hard deadline
• Non-linear curves: Reflect real-world contract penalty clauses
• Enables the auction to optimize for weighted tardiness rather than simple makespan

Instead of single-round sealed bids, a descending clock announces decreasing prices for time slots. Agents exit when the price falls below their reservation value, enabling price discovery in real-time.

• Avoids the computational burden of solving the full WDP upfront
• Provides transparency—all participants observe the clearing process
• Well-suited for perishable capacity like Just-in-Time (JIT) delivery windows
• Terminates when demand equals available capacity

Winning bids are not final. If a disruption occurs, the system triggers compensating auctions where affected agents bid to release their slots or claim new ones, forming a chain of adjustments.

• Implements the Saga Pattern for distributed manufacturing transactions
• Agents that voluntarily yield capacity receive VCG-based compensation
• Prevents deadlock where rigid schedules cannot adapt to machine breakdowns
• Maintains global schedule feasibility without centralized replanning

Agents operate under artificial budgets representing their organizational cost centers or carbon allowances. This prevents a single high-priority job from starving all others.

• Implements mechanism design with financial constraints
• Agents must solve a knapsack problem: which auctions to bid in given limited funds
• Enables multi-objective optimization balancing profit, on-time delivery, and sustainability
• Budgets can be dynamically replenished based on factory throughput

The foundational task-sharing negotiation protocol upon which most auction-based scheduling systems are built. A manager agent announces a production task with a specified eligibility specification, and potential contractor agents evaluate their capability and current capacity before submitting bids. The manager then evaluates proposals and awards the contract to the most suitable agent.

• Announcement phase: Manager broadcasts task requirements to all agents
• Bidding phase: Agents respond with cost estimates and completion timelines
• Award phase: Manager selects the winning bid and establishes a temporary contract
• Used extensively in flexible manufacturing systems and AGV dispatching

A procurement mechanism that allows bidding agents to place all-or-nothing offers on bundles of heterogeneous resources rather than individual time slots. This captures synergistic value where the utility of winning a set of resources together exceeds the sum of winning them separately.

• Bidders express complementarity between resources (e.g., a milling machine plus a specific cutting tool)
• Prevents the exposure problem where an agent wins only part of a needed bundle
• Winner determination becomes an NP-hard optimization problem requiring specialized solvers
• Common in transportation lane procurement and multi-machine production cell allocation

A sealed-bid combinatorial auction mechanism designed to incentivize truthful bidding as a dominant strategy. Each winning agent pays the marginal harm their win imposes on other participants, calculated as the difference between the optimal total value with and without that agent's participation.

• Eliminates strategic bid shading; agents bid their true private valuation
• Payment rule: winner pays the externality they impose on others
• Computationally intensive for large combinatorial spaces
• Applied in scarce resource allocation where truthful preference revelation is critical for global efficiency

The inverse game theory discipline of designing auction rules and incentive structures so that self-interested, rational agents are motivated to reveal truthful private information, resulting in a globally optimal outcome despite decentralized decision-making.

• Incentive compatibility: Rules ensure honest bidding is the best strategy
• Individual rationality: Agents voluntarily participate because expected utility is non-negative
• Allocative efficiency: Resources flow to the agents that value them most highly
• Critical for preventing gaming behavior in multi-agent production scheduling systems

The algorithmic preprocessing step that analyzes and orders manufacturing tasks based on prerequisite constraints before auctioning time slots. This ensures that a winning bid for a downstream operation does not create a work-in-process starvation scenario where upstream tasks remain unassigned.

• Constructs a directed acyclic graph (DAG) of task dependencies
• Topological sorting determines the eligible auction sequence
• Prevents deadlock where two agents each hold resources the other requires
• Integrates with Just-in-Time Sequencing (JIT) to minimize buffer inventory between auctioned operations

A mathematical framework that formalizes auction-based scheduling as a set of variables (time slots, machines), domains (feasible assignments), and constraints (due dates, precedence, capacity). The auction mechanism acts as a distributed solver, with agents bidding to satisfy subsets of constraints.

• Hard constraints: Physical impossibilities like machine capacity violations
• Soft constraints: Preferences like minimizing makespan or prioritizing high-margin orders
• Agent bids encode their ability to satisfy constraints at specific cost levels
• Backtracking and constraint propagation resolve conflicts when multiple winning bids are incompatible