Throughout evolution the brain has optimized the control of the body across species, and in order to understand the relationship between brain and body it is imperative that we model the sensorimotor transformations underlying embodied control. As part of a larger coordinated effort, we are in the process of developing a general-purpose behavioral platform for simulating biomechanics, behavioral dynamics, and the neural circuits underlying embodied control. The appropriate level of abstraction for modeling body actuation is highly debated and each actuator type comes with associated tradeoffs. A central question of this research is to investigate the appropriate level of abstraction for biomechanical actuation during complex motor control tasks. Specifically, we investigate the impact of using torque and position actuators, like those found in robots, in contrast to biologically-inspired muscle actuators in neural control models of the mouse forelimb. We implement a reinforcement learning framework to perform a dextrous forelimb reaching task in a simulated physics environment. We present results that indicate that neural embodied control policies may develop different neural representations depending on the biomechanics. This work provides evidence to support the careful use of biomechanical actuators in embodied NeuroAI modeling.