Autonomous agents must accomplish complex tasks while interacting with uncertain opponents and coordinating with teammates. This dissertation develops methods for robust and adaptive decision-making through opponent modeling, adversarial policy synthesis, and learning guided by temporal logic task specifications. It also introduces a method that attributes shared safety costs to individual agents to support safe cooperation. Experiments in autonomous racing, robotic control, cooperative transport, and networked systems demonstrate progress toward capable agents that adapt to others and coordinate under shared safety requirements.