The nonlinearity and uncertainty of machine parameters impose great difficulties in accurate modeling and optimal efficiency control of interior permanent magnet synchronous motors (IPMSMs) drives. The goal of this thesis is to propose novel control schemes to achieve accurate and robust optimal efficiency control of IPMSM drives in both constant torque region and field weakening region. Firstly, this thesis proposes a novel virtual signal injection (VSI) based control method for maximum torque per ampere (MTPA) operation and voltage constraint maximum torque per ampere (VCMTPA) operation of IPMSM drives in constant torque region and field weakening region, respectively. The proposed method injects a small virtual current angle signal mathematically for tracking the MTPA/VCMTPA operating points and automatically generates optimal current commands by utilizing the inherent characteristic of the MTPA/VCMTPA operations. Secondly, this thesis proposes a novel concept that utilizes rotor synchronous reference (d-q) frame based searching techniques to compensate the MTPA/VCMTPA control errors of control schemes in stator flux linkage synchronous reference (f-t) frame. Without loss of generality, the proposed virtual signal injection control is adopted as an example of searching schemes in the d-q frame and the existing direct flux vector control is adopted in the thesis as an example of f-t frame based control schemes. Thirdly, this thesis proposes a novel self-learning control (SLC) scheme for MTPA and VCMTPA operations based on the proposed virtual signal injection. This control scheme can be trained online and automatically adapt to machine parameter variations. Finally, a novel hybrid control concept which combines the conventional field orientated control (FOC) and direct flux vector control (DFVC) is proposed to inherit the advantages of d-q frame based control schemes and f-t frame based control schemes.