Auctions – or bidding mechanisms – provide a natural way to distribute shared resources fairly among multiple competing agents. But how can we design autonomous agents that must bid strategically to achieve their objectives in such shared environments? In this talk, I will introduce bidding games, a simple yet expressive model that has been used for studying bidding-based decision-making over graphical arenas.

Building on this model, I will present our modular decision-making framework based on bidding games. This framework addresses problems where an agent must satisfy multiple objectives simultaneously. For instance, consider a robot that must reach a target location (objective #1) while avoiding battery depletion (objective #2). Instead of synthesizing a single, monolithic strategy that satisfies all objectives at once, our framework constructs local, bidding-enabled strategies for individual objectives. These strategies then interact through a bidding-based composition that ensures all objectives are collectively satisfied.

Much like modular paradigms in software or hardware design, our approach lowers the design and maintenance complexity of decision-making systems: local strategies can be developed, refined, and replaced independently. To the best of our knowledge, this is the first modular solution for multi-objective decision-making based on bidding principles.

The talk will be based on the content of the following paper: Avni, G., Mallik, K. and Sadhukhan, S. Auction-based scheduling. In TACAS 2024.