The main differences between single-turn and multi-turn absolute value encoders lie in how they handle position information
Single-turn absolute value encoders and multi-turn absolute value encoders are two common types of position encoders used to encode input position information into vector representations, each operating with distinct principles.
Single-Turn Absolute Value Encoder：
The single-turn absolute value encoder uses a scaled angle value to represent specific positions. Typically, it maps positions within a fixed range onto a continuous domain. For example, it can map positions from 0 to 360 degrees into a continuous range from 0 to 1. The primary advantage of this encoder is its simplicity and ease of implementation. However, it can only represent positions within a single revolution and cannot distinguish between multiple positions at the same angle.
Multi-Turn Absolute Value Encoder：
The multi-turn absolute value encoder uses extra data to handle multiple revolutions. It applies modulo operations to find complete rotations and uses a single-turn encoder to show positions within each rotation. For example, it might use a binary vector to show rotations and map the current position from 0 to 1, combined with the binary rotation vector. This allows telling apart positions with the same angle and gives a thorough position representation.
In summary, a single-turn absolute value encoder is suitable for representing positions within a single revolution, while a multi-turn absolute value encoder is appropriate for distinguishing positions across multiple revolutions. The key to choosing the encoder lies in understanding the positional requirements and the application context of the model.
