CollaborativeAI is a research-oriented automated negotiation framework built using NegMAS.
The project implements and evaluates multiple negotiating agents in a shared discrete bargaining domain, allowing direct comparison of different negotiation strategies.
This repository includes:
- Multiple custom negotiation agents (MiCRO-style, time-based, reactive variants)
- A centralized match runner (
run_match.py) for controlled experiments - Configurable negotiation parameters (deadline, concession power, reservation values)
- Linear utility functions for price-based negotiation
- Built-in visualization of negotiation traces and Pareto outcomes
The primary objective is to experimentally evaluate how different agent strategies perform against each other under identical negotiation conditions.
Current experiments use a single-issue price negotiation:
- Price range:
0–99 - Buyer prefers lower prices
- Seller prefers higher prices
- Linear utility functions
- Discrete SAO (Stacked Alternating Offers) protocol
Utilities are private to each agent, consistent with standard negotiation assumptions.
- Full outcome enumeration
- Utility-sorted outcome space
- Time-based aspiration function (Boulware-like concession)
- Concession driven by relative time
- Adjustable concession power parameter
- Concede only when opponent improves
- Optional late-stage time pressure mechanisms
All agents can be matched directly against each other.
To execute a negotiation match:
python run_match.py