One step beyond the abstract.

Before grasping, location matters. Afterward, the question changes to what is happening within the grasp. Imagine moving a transparent cup: seeing it does not by itself reveal whether contact is secure or the cup is beginning to tilt. A useful reading starts by locating the missing information at each stage of an action. That gives a tactile sensor a specific role rather than treating it as an extra sense added for its own sake.

The review covers mechanics-based interpretation and learning representations or policies from data, including imitation learning, reinforcement learning, VLA and predictive models. An editorial way to organize the material is to separate sensor input, internal representation and resulting action. Ask which connection a particular method studies. The two perspectives need not be read as completely replacing one another.

A review maps approaches; it is not a ranking from one shared experiment. An editorial comparison should record the observed quantity, objects, action adjustment and completion criterion together. A clearer signal and a better task outcome are also different findings. Look for the connection showing how improved observation changed what the robot could do.

Fetching something across a room sounds like one task, but it joins approach position, body posture and reaching. Separate videos of walking or stationary grasping do not describe that entire chain. A useful reading of LocoManiLab begins at the connection between movement and manipulation: reaching the place from which a hand action can actually work.

LocoManiLab includes 234 tasks, five household scene categories and 51.5 hours of simulated teleoperation. It compares twelve policies with a shared controller and separately studies controller variation. An editorial reading should mark what remains fixed and what changes. That helps separate a policy's contribution from the machinery translating its choices into movement.

The 51.5 hours describe simulated teleoperation data, not autonomous service in homes. Task count likewise counts test items, not successes. With those distinctions in place, the benchmark's value becomes clearer: it provides common material for comparing approaches and locating failure. Attach each number to what it measures rather than turning every large number into a deployment claim.

A table supports an object during a grasp; contact with a floating target can alter its motion. Compare an imagined ball on a table with one slowly rotating without support. Shape may be unchanged, while approach timing and the response after contact become different problems. Free-floating capture asks whether familiar manipulation skills survive that change.

AstroBench studies free-floating manipulation with a fixed-base arm and separates initial motion conditions and task-target combinations. The fixed base does not imply a fixed target. Marking what moves and what stays anchored distinguishes this setup from one where the entire robot floats. That mechanical distinction is more informative than the spacecraft imagery around the scene.

The counts belong to selected policies and recorded observation-action settings. They do not describe every space robot or an operational mission's probability of success. Comparable inputs, targets and completion criteria matter. Those details make improvement explainable: another researcher needs to know what changed and what remained fixed.