Robot Action-Space Mapping¶
Modeling method — instantiates Affordance Shaping
Maps the actions a robot can actually execute in its environment — reachable, collision-free, within its own limits — so the intended action lies inside the feasible space and the harmful ones fall outside it.
Most affordance-shaping mechanisms arrange the world for people; Robot Action-Space Mapping does it for a machine, and it does it by building an explicit model first. It treats the robot's action space not as the list of tasks its designers imagined but as the intersection of the robot's own body — reach, degrees of freedom, payload, sensor field — with the geometry of the world it stands in. Its defining move is to draw the boundary between the feasible region and the forbidden one (self-collision, joint and torque limits, keep-out zones, sensor blind spots) explicitly, so that the intended action can be checked to lie inside the feasible set and the harmful actions can be walled outside it. The output is not a description but a queryable substrate a planner consults at run time.
Example¶
A warehouse team is about to trust a six-axis arm to pick items out of shelf bins. Before it moves for real, they map its action space. For each bin, they compute which grasp poses are actually reachable without the elbow clipping the shelf, which stay within payload, and which would drive the wrist past its joint limit. They build it as a configuration-space model: the shelving, neighbouring bins, and the safety fence become forbidden regions; the free region is the set of collision-free arm configurations.[n1]
The map surfaces a trap: the top-back corner of every bin is technically reachable, but only through a pose whose approach clips the shelf lip. That reachable-but-unsafe pocket is exactly the kind of thing a task list never shows. So they reshape the affordance — shallower bins, or the planner excludes that region — until the intended action (pick anything in the bin) sits wholly inside the mapped-feasible space and the collision path is out of bounds.
How it works¶
- Represent actions as regions, not a to-do list. Poses and motions become points and volumes in a configuration or action space that can be searched, not bullet items to be enumerated by hand.
- Intersect body with world. The robot's kinematic and sensing envelope is combined with the environment geometry to yield the feasible set — what is reachable here, for this machine.
- Demarcate the forbidden boundary. Collisions, joint and torque limits, keep-out zones, and blind spots are marked as an explicit surface bounding the feasible region.
- Stay queryable. The map's purpose is to be consulted by a motion planner at run time, so it is built to be executable, not merely read once.
Tuning parameters¶
- Resolution — how finely the action space is sampled. Finer sampling catches narrow feasible passages but costs computation and memory.
- Safety margin — how far obstacles are inflated. Wider margins are safer but shrink the feasible set and can declare a genuinely needed action infeasible.
- Modeled dynamics — static reach only, or velocity, torque, and compliance too. More physics makes the map truer and much costlier.
- Static vs. live — computed once offline, or continually updated from live perception as the scene changes.
- Feasible vs. preferred — whether the map only splits feasible from forbidden or also scores actions by cost and robustness.
When it helps, and when it misleads¶
Its strength is turning "the robot should be able to do X" into a checkable claim, and surfacing reachable-but-unsafe traps before the arm swings — giving a planner a searchable substrate instead of a wish. It is the difference between hoping a motion is safe and being able to ask.
Its central failure mode is that the map is a model, and the feasible set it reports can drift from reality — the sim-to-real gap. An unmodeled cable, a compliant object, sensor noise, or a person stepping in makes a "feasible" action fail or a "forbidden" cell actually fine. The classic misuse is trusting the offline map as ground truth and skipping live validation — or, worse, quietly shrinking the safety margin until a motion someone already committed to comes out feasible. The discipline that guards against this is to validate mapped-feasible actions against real execution, keep honest margins, and refresh the map whenever the environment moves.
How it implements the components¶
Robot Action-Space Mapping fills the spatial, feasibility side of the archetype's machinery — the parts a formal action-space model can produce:
action_possibility_inventory— its core output: the enumerated set of actions actually executable here, as a searchable region rather than a hand list.environment_substrate_map— the workspace geometry, obstacles, and surfaces the actions play out against, in the same coordinate frame as the robot.misuse_and_trap_boundary— the explicit forbidden surface (collisions, limits, keep-out zones) that bounds the feasible set and names the reachable-but-unsafe traps.
It does not model the agent's task-level capabilities or the goal-versus-harm intent — that framing is Task and Capability Analysis — and it designs no perceivable cue, since a machine reads geometry, not signifiers; the signifier layer belongs to Signifier Prototyping.
Related¶
- Instantiates: Affordance Shaping — it maps and then reshapes the affordances available to a robot.
- Consumes: a capability model of the machine (reach, limits, sensing) supplied by Task and Capability Analysis, applied to the robot.
- Sibling mechanisms: Task and Capability Analysis · Usability or Field Test · Signifier Prototyping · Safe Default or Preselected Path · Wayfinding Marker · Affordance Audit · Physical or Digital Keying · Friction Adjustment · Prototype A/B or Multivariate Test
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Robot Action-Space Mapping operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it maps the actions a robot can actually execute in its environment — reachable, collision-free, within its own limits — so the intended action lies inside the feasible space and the harmful ones fall outside it.
Independent corroboration: The frozen evidence defines Robot Action-Space Mapping as 'Maps the actions a robot can actually execute in its environment — reachable, collision-free, within its own limits — so the intended action lies inside the feasible space and the harmful ones fall outside it', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Robotics & Automation
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Specialized
Rationale: Reachability, collision constraints, and actuator limits define a robot's feasible action space in robotics.
Related originating lineages:
- Computer Science & Software Engineering — Motion planning supplies algorithmic action-space representation.
- Engineering & Design — Mechanical and control engineering materially determine physical feasibility.
Review resolution: Both blind reviewers agree that robotics_automation is the primary historical origin. Explicit reconciliation of origin mode disagreement, encyclopedia synthesis disagreement starts from reviewer_a’s mechanism-specific evidence: Reachability, collision constraints, and actuator limits define a robot's feasible action space in robotics. Reviewer A proposed alternates=computer_science, engineering_design, origin_mode=single_lineage, domain_reach=specialized, and encyclopedia_synthesis=false; reviewer B proposed alternates=computer_science, engineering_design, origin_mode=cross_disciplinary_synthesis, domain_reach=specialized, and encyclopedia_synthesis=true. The final record retains every independently supported alternate from either review (computer_science, engineering_design) without an arbitrary cap, selects origin_mode=cross_disciplinary_synthesis to represent the combined lineage evidence, and keeps domain_reach=specialized and encyclopedia_synthesis=false from the more mechanism-specific assessment. Present-day transfer is recorded as reach and is not treated as proof of historical origin.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The map is a model, not the territory: its authority ends exactly where perception and dynamics exceed what it captured. That is why it pairs with a live-perception loop and real-world execution checks — the machine-facing counterpart of the Usability or Field Test that validates a human affordance against actual behaviour.
[n1] In robotics the configuration space (C-space) represents each pose of the robot as a single point; collision-free motions form the free region and obstacles and limits form forbidden regions. Lozano-Pérez's formulation is the standard basis for motion planning, and framing the action space this way is what makes "feasible vs. forbidden" a boundary you can compute against. ↩