Recent effort in exoskeleton control resulted in reduction of human metabolic consumption during ground-level walking. In this context, solutions that would enable biomechanical and metabolic benefits across large repertoires of motor tasks would be central in supporting the human in both medical and industrial scenarios. With this idea in mind we created a muscle-driven controller based on electromyography (EMG)-driven musculoskeletal modeling that we interfaced with the robotic bi-lateral Achilles ankle exoskeleton previously developed in our group. Preliminary results on one healthy individual show the possibility of continuously decoding EMG-dependent muscle force and resulting ankle joint moment patterns in real-time across a range of diverse motor tasks. We demonstrate that this information can be used to establish a human-exoskeleton interface with high-resolution at the level of single muscle mechanics.