Swarm robotics is a new and promising approach to the design and control of multi-agent robotic systems. In this paper we use a model for a second order non-linear system of self-propelled agents interacting via pair-wise attractive and repulsive potentials. We propose a new potential field method using dynamic agent internal states to successfully solve a reactive path planning problem. The path planning problem cannot be solved using static potential fields due to local minima formation, but can be solved by allowing the agent internal states to manipulate the potential field. Simulation results demonstrate the ability of a single agent to perform reactive problem solving effectively, as well as the ability of a swarm of agents to perform problem solving using the collective behaviour of the entire swarm.