Creating grippers that can perform tasks similar to a human hand's capability is a challenge yet to be solved. Most electrical actuators used in grippers have been geared, working at high numbers of revolutions per minute. By being geared, the motor can achieve high precision and efficiency in repetitive tasks. However, while efficiency and repeatability may be optimal in industrial circumstances, actuators operating in human and natural environments may require different properties, properties where high torque, speed, and relative low momentum are more desirable than exact position. Effectively picking up an egg from a table or sorting laundry from a washing machine may be such a task. Tasks that geared grippers struggle with without increased complexity in the form of tactile sensing. Unfortunately, incorporating touch sensing into robotic grasping has proved challenging due to sensor sensitivity, cost, and difficulties of integrating tactile inputs into standard control schemes. By changing to direct-drive motors, touch sensing might be achievable without tactile sensors and transmission complexity, increasing the accessibility of affordable prosthetics and robotic grippers with human-like performance. To find out what direct-drive motors are capable of, this thesis aims to 1) create and control a direct-drive actuated mechanical finger 2) use position sensors and current feedback from a direct-drive motor to find out if, and to which degree, it is possible to detect and provided loss of grip. The results show that a direct-drive actuated finger controlled by field-oriented control can be used to detect low contact forces using an exponential average filter over the torque/current measurements. It is possible to detect slip, squeeze, slide, and stable grasp by looking at torque/current data. Detecting slides will require finger vibration if slides do not cause a rotation on the motor. The vibration causing the torque/current to change and be detected may be too large to be used in any applications where oscillations on both object and finger and increased noise is unwanted. A ResNet18 classifier has been implemented and can distinguish between stable grasp, slip, and squeezing with an accuracy of 89% based on the current drawback data from the motor.