Relative and Absolute modes of KUKA operationOperating KUKA in the relative mode

First: KRL programmed motions, and RSI controller motions, are completely separate things, although they do share the same concepts.

KRL motion locations are normally programmed as E6POS variables, which are Cartesian coordinates, although some positions, like HOME, are usually programmed in E6AXIS, to explicitly control the axis angles.

Any Cartesian location in KRL is treated as a multi-stage 6-DOF transform, using the current values of $BASE and $TOOL. This can become rather complex, but briefly, $TOOL defines the location of the Tool Center Point (TCP) whose position the robot is controlling, relative to the mounting flange on Axis 6. $BASE defines that current active reference frame of the robot, as a transform relative to $WORLD. $WORLD is usually colocated with $ROBROOT, which is located in the bottom center of the robot's baseplate.

Due to the interaction of $TOOL, $BASE, and E6POS positions, a single Position can move the robot to multiple locations in physical space, depending on changes made to $BASE and/or $TOOL. A palletizing program, for example, might shift $BASE as the counters increment through the palletizing pattern, while an end effector mounting both a vision system and drilling tool might switch between $TOOL settings at the same programmed Position, in order to bring both tools to bear on the same point in space without maintaining separate copies.

Relative motions can be commanded in either Cartesian or Joint space. Cartesian relative motions can also be commanded relative to $BASE, or $TOOL.

That's... some pretty deep water, there. You really want to be well-grounded in how KUKAbots function when you try this -- RSI is powerful, but also dangerous. It can cause the robot to move at near-automatic speeds even in T1, and a badly-tuned PID object can make the robot move so violently as to damage itself just from inertia. RSI operates "beneath" several of the "safety nets" in the system.

POSCORR objects generally operate in relative mode already, starting from the robot's position when the RSI_ON command was issued. The POSCORR objects can be selected to operate in Tool or Base. However, these are explicitly positional commands -- to command velocity, you're essentially going to have to integrate your desired velocity into an equivalent positional increment. Since RSI (usually) updates every 12ms (4ms for some limited modes), you can work backwards to your integral factor from there. You can build a set of PID objects between your input and the POSCORR to do the conversion.