In this paper, we focus on planning credible walking paths in real-time for a potentially highly congested crowd of au- tonomous pedestrians. For this purpose, we exploit the prin- ciple of least effort, applied to human navigation, which pos- tulates that credible behaviours emerge as a function of the organism's propensity to minimize metabolic energy expen- diture with respect to task, environment dynamics, and or- ganism's constraints to action. We therefore propose a consistent problem formulation for the navigation task where both individual and collective dynamics are taken into account. Each pedestrian is represented as a situated agent who tries to reach its destination by following energy efficient paths. Agents are autonomous, and at the same time, sub- ject to the environment dynamics. They interact with each other through the environment in order to estimate their en- ergy expenditure relatively to their tasks. Our formulation results in a generic and scalable multi-agent model, capable of simulating individual and collective behaviours regardless of the number of agents.