Huber Loss
This is the loss function used in the original Deep Q-Learning paper for Atari Breakout.
The loss is around the center looks like a parabola (squared loss), but as it goes outside, it becomes linear.
Huber Cost Function is also used in Bundle Adjustment.