​Powered prostheses help transfemoral amputees regain their walking mobility. However, currently available powered leg prostheses provide limited hours of continues walking on a single battery charge. Thus, we designed an optimal low-level control system for transfemoral prostheses with energy regeneration, which allows an amputee to walk similarly to able-bodied persons. Our contribution was to capture net surplus energy from a prosthetic knee to alleviate the power requirement of a prosthetic ankle. This work mimicked the biological transfer of energy using ultracapacitors instead of ligaments and tendons. We modeled a prosthesis test robot with energy regenerative electronics. We demonstrated the significant effect of ground reaction force (GRF) on energy regeneration characteristics using an energy balance equation. We therefore designed an optimal passivity-based tracking/impedance control to impose a desired relationship between GRF and deviation from reference trajectories. The global asymptotic stability was proven using Lyapunov approach.  We showed that trajectory tracking and energy regeneration are conflicting objectives. We thus used constrained multi-objective optimization (MOO), nondominated sorting biogeography-based optimization, to find optimal impedance control parameters to compromise between the two objectives. A tradeoff solution resulted in decent knee angle tracking error (0.0178 rad) and decent energy regeneration (131.2 J) in a walking robotic experiment. Our experimental results support the possibility of net energy regeneration at the semi-active knee joint with human-like tracking performance. The results indicate that advanced control and optimization of ultracapacitor-based systems can significantly reduce power requirements in transfemoral prostheses.