Autonomous adaptive agents acting in a real or simulated environment exhibit behavioral strategies governed by complex dynamics. In order to have a mathematical description of the behavior of a population of robots, their position is recorded while performing repetitive actions in interaction with the world. The time series is analyzed within the framework of dynamical system theory, using several nonlinear tests to confirm the presence of chaos, including nonlinear prediction error for determinism, cross-prediction error for stationarity and surrogate data testing on time series where nonlinearities have been disrupted. Results show that robots characterized by a chaotic dynamic have a better ability to adapt to known and novel environments. Furthermore, Lyapunov exponent and fractal dimension are correlated with the overall performance of the robots.