<p>Heterogeneous dynamics in multi-agent dynamical systems (MADS) can lead to a big challenge in achieving desired consensus or coordinated behavior because the agents respond differently under the same environment or same input. This can be even more challenging when the agents’ dynamics is highly-nonlinear and completely-uncertain. This paper proposes a Deterministic Learning (DL) based scheme for the joint problems of heterogeneous dynamics learning and formation control for a class of discrete-time MADS, which are allowed with heterogeneous, highly-nonlinear, completely-uncertain dynamics. A key contribution of this scheme is its capability of accurately learning the systems’ heterogeneous dynamics through the tracking control with provable stability and convergence of the overall closed-loop system. This scheme is designed under a virtual-leader-following-based dynamics learning framework, consisting of virtual-leader-based distributed adaptive observers and DL-based decentralized dynamics learning controllers. Empowered with the learning architecture, the scheme can develop distinctive capabilities of: (i) achieving locally-accurate identification/learning for MADS’s dynamics with provable convergence of associated parameters to their optimal values; (ii) accomplishing dynamics-knowledge acquisition and reutilization to online improve control performance and computational efficiency for MADS; and (iii) realizing formation tracking control for each individual agent via its separate dynamics learning in a fully-distributed manner without using global information. Extensive simulation study is conducted to demonstrate the effectiveness and advantages of our approaches.</p>