Skip to content

Latest commit

 

History

History
509 lines (381 loc) · 34.6 KB

File metadata and controls

509 lines (381 loc) · 34.6 KB

DimOS Robotics Context

DimOS describes robots, actuators, control surfaces, and manipulation planning using precise robotics terminology. This glossary records stable domain language only, not implementation details.

Language

Planning group: A named subset of a robot model's joints and frames that can be selected as a planning unit. Avoid: move group, joint group

Composite planning group: A planning group that represents coordinated motion across multiple selected planning groups. Avoid: group combination, combined groups, multi-group plan

Auxiliary planning group: A planning group included in a planning request that may move as part of the plan but does not have its own task-space target in that plan. Avoid: extra group, passive target group, unconstrained target

Linear TCP path: A motion recipe where the tool center point follows a straight Cartesian segment from its start pose to its target pose within configured Cartesian tolerance. Avoid: linear joint motion, linear motion

TCP target planning: A manipulation-planning capability where the final tool center point pose is constrained but the intermediate path shape is not. Avoid: Cartesian planning when path shape is unspecified, linear TCP path planning

Cartesian servo / IK control: A live manipulation-control capability that follows Cartesian commands through inverse kinematics during execution rather than producing an offline geometric path. Avoid: Cartesian planning, linear TCP path planning, TCP target planning

Linear TCP trajectory smoothing: A manipulation-planning capability that makes a Linear TCP path executable without treating every intermediate Cartesian sample as a stop, while preserving Cartesian-line tolerance. Avoid: waypoint skipping, making linear motion faster

Adaptive-conservative smoothing: A trajectory smoothing policy that starts with an aggressive simplification and, on validation failure, preserves more of the original path rather than relaxing correctness tolerances. Avoid: tolerance loosening, unsafe smoothing retry

Trajectory post-processing pipeline: A staged manipulation-planning capability that may refine a geometric path, validate the refinement, assign timing, and apply execution-oriented smoothing while preserving the path's declared constraints. Avoid: hidden waypoint hack, one-off retiming step

Path-constraint metadata: Optional planning metadata attached to a geometric path that declares the path constraints any post-processing must preserve. Avoid: planner debug data, visualization-only metadata

Non-blocking smoothing fallback: A trajectory post-processing policy where smoothing failures fall back to the original geometric path rather than failing parametrization; the worst expected outcome is a slower valid trajectory. Avoid: strict smoothing gate, smoothing-required parametrization

Composite RoboPlan model: A RoboPlan-facing robot model that represents multiple registered robot models as one planning scene. Avoid: combined URDF, merged robot scene

Planning world: The authoritative belief state for manipulation planning, including robot state and scene state used by planners. Avoid: planner context, backend instance

Planning-scene synchronization: The capability that keeps the Planning world's obstacle state consistent with selected authoritative scene sources. Avoid: environment sync, world sync, simulator sync

Visualization scene mirroring: The non-authoritative capability that renders Scene Registry state or Planning world projections for user inspection without serving as the planner's safety source. Avoid: planning-scene synchronization when referring only to display updates, visualization sync as a planner guarantee

Scene Registry: The authoritative catalog of scene entities known to DimOS across perception, simulation, and explicit declarations. Avoid: object registration module, obstacle monitor, planning world

Scene entity: A spatial item represented in the Scene Registry, including physical objects, simulator-authored geometry, declared regions, dense maps, pointcloud maps, voxel maps, or other environment representations. A Scene entity does not require semantic object meaning; downstream modules decide how to interpret and project it. Avoid: obstacle when referring to the registry-level item, registered object when the source is not perception

Scene payload: A typed data payload carried inline by a Scene entity, such as primitive geometry, mesh geometry, object bounds, pointcloud data, voxel map, occupancy grid, TSDF, or surface graph. Downstream modules inspect payload type to decide whether and how they can consume the Scene entity. Avoid: semantic type when referring to the spatial data form, obstacle type when the entity is not specifically a planning obstacle, representation when referring to the concrete carried data, external payload references in v1

Camera-pose odometry: A pose stream that reports a depth or lidar sensor frame's pose in the mapping frame for spatial mapping. For wrist-mounted cameras, Camera-pose odometry is derived from robot forward kinematics plus the camera extrinsic transform; it is not necessarily mobile-base odometry. Avoid: robot odometry when referring specifically to the sensor pose needed for pointcloud integration

Self-filtering: The removal of a robot's own body, gripper, sensor mount, or other ego geometry from perceived scene data before that data is treated as environment belief. Avoid: obstacle filtering when the filtered geometry belongs to the observing robot itself

Scene source: A producer that asserts Scene entities into the Scene Registry, such as perception, simulation, or explicit operator declarations. Avoid: obstacle source when the producer is not planner-specific, perception source when referring to all scene producers

Planning projection: The planner-facing spatial representation of a Scene entity, which may be simplified, inflated, omitted, or otherwise adapted for collision checking. Avoid: object geometry when referring to planner-specific collision geometry, source geometry

Scene source snapshot: A complete statement of the Scene entities currently asserted by one scene source, used to reconcile the Scene Registry with that source's latest authoritative state. Avoid: global scene dump, planning-world snapshot

Scene entity delta: An incremental add, update, or remove assertion for one Scene entity in the Scene Registry. Avoid: obstacle command when the change is registry-level, collision object message

Source-scoped identity: A Scene entity identity formed from the scene source and that source's stable local identifier, without assuming entities from different sources refer to the same physical thing. Avoid: global object id, automatic semantic merge

Trajectory parametrization: The manipulation-planning capability that assigns time to a geometric joint path under motion constraints. Avoid: trajectory generation, retiming when referring to the broader capability

Generated trajectory: A manipulation-planning artifact that represents a geometric path after trajectory parametrization, ready for preview, validation, benchmarking, or execution planning. Avoid: timed generated plan, generated plan when referring to the time-parametrized artifact, joint trajectory when referring to the manipulation-level artifact

Shared trajectory time domain: The single timing basis a generated trajectory uses for all selected joints and robot-local projections in a composite or multi-robot motion. Avoid: independent per-robot timing, per-arm retiming when referring to coordinated composite motion

Trajectory dispatch: An execution-preparation artifact that derives control-task-specific joint trajectory messages from a generated trajectory without changing the generated trajectory's canonical global timing. Avoid: generated trajectory projection, execution-time parametrization, per-task generated trajectory

Robokin kinematics backend: A DimOS kinematics backend that presents multiple robokin-supported inverse-kinematics engines through one robotics-facing capability. Avoid: Oink backend, RoboKin world backend, single-engine Oink solver

Robokin engine: A specific solver implementation selected inside the Robokin kinematics backend, such as Placo, Pyroki, or Oink. Avoid: Robokin backend, world backend, planner backend

RoboPlan kinematics backend: A DimOS kinematics backend implemented by RoboPlanWorld using RoboPlan-backed model state, planning groups, Jacobians, and collision state. Avoid: Robokin backend, separate RoboPlan IK world, planner-only RoboPlan integration

RoboPlan Oink IK solver: RoboPlan's task-based inverse-kinematics capability for solving one or more frame pose targets under joint constraints. Avoid: hand-written Jacobian IK, Robokin-only Oink wrapper

Coordinated simulation clock: A simulation benchmark clock policy where simulator time advances in lockstep with the DimOS control coordinator clock. Avoid: autonomous simulator loop, write-triggered stepping, settle step

Runtime sidecar: A separate process or environment that owns a benchmark simulator backend while DimOS owns orchestration, control integration, skills, and artifacts. Avoid: plugin, embedded simulator

HTTP runtime removal gate: A migration success condition requiring existing HTTP runtime sidecar servers, clients, payload fetch endpoints, and HTTP-first demos to be removed once their behavior is covered by Simulator Runtime Modules. Avoid: HTTP fallback, target simulator architecture, long-term transport boundary

Simulator Runtime Module: A first-class DimOS Module that represents a simulator runtime at the blueprint boundary while preserving simulator ownership of benchmark reset, stepping, observations, and scoring. Avoid: HTTP runtime sidecar, benchmark script launcher, embedded simulator

Simulator runtime blueprint helper: A package-local blueprint factory that registers a simulator runtime's named Python project environment and places its Simulator Runtime Module into that environment using the standard blueprint placement API. Avoid: module-local deployment flag, global sidecar registry, caller-written placement boilerplate

Remote module worker: A separate Python environment that hosts first-class DimOS Modules while the main DimOS process owns blueprint orchestration and module coordination. Avoid: arbitrary sidecar service, embedded optional dependency

Venv module worker: A same-machine Python virtual environment that hosts first-class DimOS Modules separately from the main DimOS Python environment. Avoid: remote deployment, sidecar service, optional dependency import guard

Module import descriptor: A portable identity for a DimOS Module class that a venv module worker can import in its own Python environment. Avoid: pickled module class, shared interpreter object, source checkout assumption

Module placement: A blueprint-level decision that chooses where a DimOS Module instance runs, such as the default worker pool or a named venv module worker. Avoid: intrinsic module type, stream transport, permanent class identity

Local worker control channel: A same-machine private channel used by the coordinator process to send lifecycle and wiring commands to a worker process. Avoid: stream data plane, remote public API, transport topic

Multiprocessing connection control channel: A local worker control channel implemented with Python's multiprocessing connection Listener/Client machinery so separately launched Python interpreters can exchange DimOS worker protocol messages. Avoid: stream transport, stdio protocol, remote deployment API

Named venv: A runtime-resolved label for a Python virtual environment that can host venv module workers. Avoid: hardcoded interpreter path in blueprint, deployment target, stream transport

Worker protocol runtime: The DimOS runtime surface that a worker environment must provide to receive coordinator lifecycle commands and host first-class Modules. Avoid: full optional dependency set, application module package, public remote API

Module contract: A lightweight coordinator-visible DimOS Module class that declares the streams, module references, RPC surface, and config expectations used for blueprint wiring. Avoid: heavy implementation module, connection-only shim, runtime sidecar

Module IO contract: The coordinator-visible set of typed stream inputs and outputs a DimOS Module exposes for blueprint wiring. Avoid: stream schema, port list, dynamic ports

Follower-limit clamp: A teleoperation safety boundary where commands derived from a leader device are constrained to the follower robot's legal joint range before execution. Sender-side clamping improves teleop behavior, while downstream control and hardware layers may still clamp defensively. Avoid: leader limit, calibration range clamp, normalization range

Gripper endpoint calibration: A leader-gripper calibration step that records the raw positions corresponding to fully open and fully closed gripper states. It is distinct from arm-joint zero calibration because gripper control is based on an open/close interval rather than a neutral joint angle. Avoid: arm range calibration, gripper zero pose, generic min/max dashboard

Leader zero pose: The designed natural pose of a teleoperation leader arm that corresponds to the follower arm's all-zero joint configuration. Capturing this pose defines the leader's zero offsets for arm-joint teleoperation. Avoid: arbitrary neutral pose, comfort pose, range calibration pose

Leader joint assignment: The calibration-level association between a semantic leader joint name and the physical motor id that supplies that joint's reading. It is used when the physical motor ordering differs from the follower joint order. Avoid: hardcoded wrist remap, runtime joint swap, follower joint alias

Startup alignment: The operator responsibility to place a teleoperation follower near the leader-implied command before enabling live authority. It prevents first-command jumps when no automatic follower-state gate is present. Avoid: calibration, homing, sender-side clamp

Visualization-only teleop test: A teleoperation validation mode where a real leader device drives commands rendered against a follower model without connecting to follower hardware or executing physical motion. "Visualization-only" describes the follower side, not the leader input. Avoid: hardware validation, fake leader demo, dry run when physical execution is possible

Real-hardware opt-in: A teleoperation bring-up boundary where follower hardware remains mocked unless the operator provides an explicit hardware connection setting. The presence of that setting means the follower may connect and physically move. Avoid: implicit hardware fallback, hidden arming flag, visualization-only mode

Follower-observed teleop visualization: A teleoperation visualization mode that renders the follower-side state reported through the control stack rather than the leader-derived command alone. It answers "what is the follower doing or reporting?" rather than only "what is the leader commanding?" Avoid: command-only visualization, leader preview, visualization-only teleop test

Configuration-resolved module IO: A module IO contract whose streams are determined from the module's final configuration before blueprint wiring. Avoid: runtime dynamic IO, late-bound ports, generated subclass IO

Module implementation descriptor: A portable identity for the concrete Module implementation class that a venv module worker imports and instantiates for a module contract. Avoid: coordinator-imported heavy class, pickled module class, alternate stream contract

Contract compatibility responsibility: The phase-1 expectation that a module implementation descriptor names a concrete Module compatible with its coordinator-visible module contract, without an explicit verifier. Avoid: schema validation requirement, subclass proof, remote API compatibility guarantee

Venv worker failure semantics: The phase-1 rule that an incompatible or failing venv module implementation fails the normal blueprint build lifecycle instead of degrading into a partial system. Avoid: optional module fallback, background retry policy, partial blueprint success

Venv module placement API: A blueprint-level mapping that runs selected import-safe Python Module classes in named runtime environments. Avoid: new Module base class, permanent class deployment attribute, global transport setting

Python venv placement: A mapping entry keyed by an import-safe Python Module class and valued by the name of a Python runtime environment. Avoid: deployment type, stream transport key, hardcoded worker process

Reserved deploy kwargs: Internal coordinator-to-worker-manager metadata carried through module deploy kwargs during early architecture spikes and stripped before Module instantiation. Avoid: user module config, public constructor argument, permanent ModuleSpec schema

Venv worker environment config: A runtime mapping from a named venv to an existing Python executable that can launch venv module workers. Avoid: automatic venv creation, package installation plan, blueprint-embedded filesystem path

Runtime environment registry: A runtime configuration map from stable environment names to environment backends that resolve interpreters, executables, command environment variables, and optional preparation steps for DimOS-managed processes. Avoid: venv-only config, blueprint-embedded machine paths, per-module ad hoc build commands

Runtime environment preparation: An explicit pre-run action that prepares only the runtime environments used by active module placements in a loaded blueprint configuration. Avoid: implicit install during blueprint run, global environment-name command, preparing unused registry entries

Python project runtime environment: A convention-driven runtime environment rooted at a Python project directory, where standard files such as pyproject.toml, uv.lock, and optionally pixi.toml determine how the worker Python environment is prepared and launched. Avoid: per-tool runtime backend taxonomy, blueprint-embedded setup command, manually enumerated manifest paths

Pixi-backed uv runtime: A Python project runtime environment where Pixi prepares the native/toolchain layer and provides the Python interpreter used to create the project-local uv .venv; worker launch uses the .venv Python with Pixi activation environment applied. Avoid: Pixi-only Python environment, coordinator Python venv, launching without native activation environment

Toolchain-mediated worker launch: A worker launch policy for convention-based Python project runtimes where DimOS invokes the project toolchain command, such as pixi run uv run --no-sync python, instead of reconstructing activation variables or launching the venv interpreter path directly. Avoid: hand-built Pixi activation env, mutating sync during blueprint run, bypassing project runtime conventions

Named runtime environment: A stable label in the runtime environment registry that modules and worker placements reference when they need a non-default execution environment. Avoid: hardcoded venv path, Nix command string as identity, deployment type

Named venv worker pool: A set of worker processes launched from the same named venv that may host multiple compatible placed Modules without mixing Modules assigned to other Python environments. Avoid: one process per Module, shared cross-venv pool, global worker pool replacement

Worker launch strategy: The mechanism used to start and connect a worker process while preserving the shared worker pool and runtime protocol architecture. Avoid: duplicated worker scheduler, separate venv-only runtime, stream transport selection

Worker process handle: A coordinator-side abstraction for a running worker process that supports module deployment, lifecycle requests, capacity accounting, and shutdown regardless of how the process was launched. Avoid: launch mechanism, worker scheduler, module implementation

Worker launcher: A strategy object that creates worker process handles for a specific Python launch environment, such as the coordinator venv or a named venv. Avoid: pool manager, module deployer, transport factory

Import-safe module file: A Python file defining a venv-deployable DimOS Module that can be imported by the coordinator environment without importing worker-only optional dependencies at module import time. Avoid: split package requirement, top-level heavy dependency import, hidden sidecar boundary

Depth observation: A depth image interpreted with its camera calibration and the pose of its camera frame at the image timestamp. Avoid: depth point cloud, raw 3D points

Grasp target: The intended physical object or bounded scene region that grasp generation should produce grasp candidates for. Avoid: correct object, target object, grasp object

Object id: A stable identifier for a registered perceived object, used when a robot command must refer to one non-ambiguous physical object. Avoid: object-ish argument, object name when identity matters

Registered object: A perceived object that has been assigned an Object id and has enough spatial metadata to be used as a Grasp target. Avoid: detection when referring to a persistent object reference

Target-masked TSDF: A grasp-generation workspace representation where observations outside the selected Grasp target are suppressed with a deliberate cushion so the target remains intact. Avoid: censored TSDF, object-only scene

Target bounds: A world-frame axis-aligned bounding region used as a rough attention area for a Grasp target before grasp generation. Avoid: perfect object geometry, grasp geometry

Grasp candidate: A proposed end-effector grasp pose for a Grasp target, optionally carrying ranking metadata such as score, width, or approach information. Avoid: executed grasp, final pick action, object pose

Pointcloud grasp generator: A grasp-generation component that consumes point cloud observations of a Grasp target, optionally with scene context, and proposes Grasp candidates. Avoid: TSDF grasp generator, robot execution controller, perception registration module

SHM runtime data plane: A local shared-memory command/state channel between a ControlCoordinator-facing hardware adapter and a DimOS simulator client module, used when high-rate motor control must cross local process boundaries without RPC. Avoid: remote sidecar protocol, public simulator API, benchmark control plane, module object sharing

Motor state projection: The hardware-facing subset of simulator state that resembles what a raw robot driver exposes: actuator positions, velocities, efforts, commands, enable state, and errors. Avoid: task observation, scene observation, evaluator state

Whole-body motor surface: A hardware control surface that treats a robot as an ordered set of motors with per-motor state and commands, independent of whether the robot is a manipulator, mobile base, or humanoid. Avoid: manipulator-only adapter, end-effector API, task action API

Runtime motor action frame: An ordered command frame addressed to a simulator runtime's declared whole-body motor surface for one benchmark step. Avoid: backend-native action vector, opaque simulator action, stream transport payload

Benchmark episode config: A backend-facing declaration of benchmark intent that names the task, robot, runtime constraints, and evaluation setup before any DimOS blueprint is launched. Avoid: hardware config, simulator config, blueprint config

Static spatial evaluation: An evaluation where an agent may reason through multiple turns but must answer a grounded spatial question from one spatial snapshot without receiving new environment data. Avoid: short-horizon spatial evaluation, one-shot agent evaluation, embodied task-success evaluation, simulator rollout

Spatial snapshot: An immutable bundle of synchronized robot observations and accumulated spatial belief available at one selected recording timestamp. It may contain state derived from earlier observations but never data from after the selected timestamp. Avoid: RGB frame when multiple spatial modalities are present, Scene source snapshot, recording

Spatial evaluation system: The complete configured agent pathway evaluated against a spatial snapshot, including its model, instructions, spatial representations, tools, and adapters. The benchmark scores this pathway as one system rather than attributing performance to individual components. Avoid: model-only score, tool-only score, component-isolated spatial evaluation

Environment-scale spatial understanding: The ability to answer grounded questions about robot orientation, navigable free space, structural regions, and their connectivity from a spatial snapshot. Object-to-object relations may serve as calibration checks but are not the primary evaluation target. Avoid: generic visual spatial reasoning, object-relation benchmark as the primary target

Map-grounded spatial evaluation: A static spatial evaluation in which the complete spatial evaluation system answers questions from a supplied immutable map. The map may preserve sensor-derived imperfections, but its production and correctness are outside the evaluated system. Avoid: SLAM evaluation, sensor-grounded spatial evaluation, real-world map-accuracy benchmark

Sensor-grounded spatial evaluation: A static spatial evaluation in which frozen lidar-derived observations are processed through the configured mapping and agent pathway, while a separate authoritative scene description is used only to generate and score questions. Mapping imperfections are therefore part of the evaluated system. Avoid: perfect-map input, oracle geometry exposed to the agent, map-grounded spatial evaluation

Full-map spatial snapshot: A spatial snapshot produced after a fixed sensor replay has covered the benchmark scene. The resulting map is immutable during questioning and may retain errors caused by controlled sensor noise and the benchmark's source mapping pipeline. Avoid: perfect map, single-scan snapshot, interactive exploration during evaluation

Canonical benchmark map: The fixed, agent-visible map supplied by a Map-grounded spatial evaluation. It may intentionally preserve errors from a sensor and mapping pipeline, but every evaluated system receives the same map and the benchmark does not score map production. Avoid: evaluation oracle, perfect map, mapper output generated during an evaluation run

Spatial evaluation noise profile: A seeded model of sensing and localization imperfections applied to observations and poses before spatial mapping. It combines physical sensing limits with temporally correlated pose drift so the resulting map errors arise through the evaluated mapping pathway. Avoid: arbitrary costmap corruption, independent pose jitter, post-mapping image noise

Evaluation robot footprint: A fixed two-dimensional collision volume representing the benchmark robot's occupied space and safety clearance. Spatial answers concern whether this complete volume can translate or rotate safely, not whether a point can pass through free space. Avoid: point robot, target-cell occupancy, abstract free-space query

Benchmark room: A dataset-annotated navigable region whose boundary is physically expressed by walls and explicit openings in the authoritative scene geometry. Open-plan semantic subdivisions, closets, stairs, and inaccessible regions do not count as Benchmark rooms. Avoid: room label without a physical boundary, every enclosed polygon, semantic zone

Spatial query marker: A neutral label placed at a location in an agent-visible spatial representation to identify a question referent without revealing room boundaries, semantics, connectivity, or the expected answer. Avoid: semantic room label, topology annotation, answer hint

Benchmark reset authority: The control-plane responsibility for establishing a new benchmark episode state before simulation time advances. Avoid: reset topic ownership, observation stream side effect, simulator auto-reset

Semantic skill benchmark episode: A benchmark episode where the agent acts through named, task-level DimOS skills while simulator backends provide reset, observation, and external scoring. Avoid: motor-control benchmark episode, raw simulator action episode, code-as-policy episode

Agentic manipulation module: A universal DimOS skill-facing module that exposes agent-appropriate manipulation capabilities by coordinating existing manipulation and control modules. Avoid: benchmark skill container, Robosuite skill module, simulator-specific manipulation API

Resolved runtime plan: The concrete DimOS launch material derived from a benchmark episode config, including hardware components, simulator connection config, observation streams, evaluator setup, and artifact routing. Avoid: benchmark intent, user-authored task config

Runtime prelaunch orchestration: The phase that starts and coordinates the simulator sidecar environment and the DimOS blueprint environment before a benchmark episode begins. Avoid: config parsing, blueprint launch, single-process startup

Runtime asset bootstrap: A deliberate preparation phase that retrieves, stages, or validates external benchmark assets before a runtime sidecar starts an episode. Avoid: implicit sidecar download, startup mutation, hidden dataset setup

Remote runtime boundary: The network-facing protocol boundary between a DimOS simulator client and a benchmark backend process that may run in another environment or on another machine. Avoid: shared memory boundary, hardware adapter boundary, in-process simulator object

Runtime protocol schema: The shared, backend-neutral message contract used on the remote runtime boundary to describe episodes, robot motor surfaces, actions, observations, scores, and artifacts. Avoid: backend SDK type, DimOS hardware adapter type, simulator object

Runtime protocol package: A lightweight installable package in the monorepo that contains only remote runtime protocol schemas, codecs, and compatibility tests so sidecars can depend on it without installing DimOS. Avoid: DimOS submodule, simulator backend package, hardware adapter package

Runtime observation stream: A simulator-derived observation exposed through DimOS's normal typed stream contracts so visualization, agents, and evaluators can consume the same observation path. Avoid: artifact-only observation, sidecar metadata, viewer shortcut

Runtime payload reference: A protocol observation field that names retrievable binary observation data, allowing step responses to carry metadata while clients fetch image, depth, or segmentation payloads separately. Avoid: inline base64 image, local file path contract, remote shared memory

Array-native observation payload: A runtime observation payload that preserves array shape, dtype, and values across the remote runtime boundary without image compression semantics. Avoid: JPEG-first image transport, display-only frame, encoded screenshot

Runtime observation module: A DimOS module that turns remote runtime observation metadata and payload references into normal typed DimOS streams. Avoid: demo script publisher, sidecar-owned DimOS stream, artifact replay

Step-synchronized observation: A runtime observation publication policy where images and camera metadata are emitted from the same episode step that produced the motor state, score metadata, and protocol trace. Avoid: independent polling frame, unsynchronized viewer feed, latest-only observation

Script-hosted runtime demo: A plain Python demo that orchestrates sidecar startup, runtime stepping, local control plumbing, and visualization without becoming a product CLI or benchmark runner. Avoid: production runner, DimOS CLI command, artifact-only smoke test

Rerun runtime demo: A script-hosted runtime demo mode that publishes simulator observations into Rerun through DimOS observation streams for visual inspection. Avoid: simulator viewer shortcut, saved-image artifact, direct Rerun-only bypass

Stream-backed runtime visualization: A visualization path where runtime observations are rendered by consumers of DimOS streams, not by calling a visualization SDK directly at the runtime boundary. Avoid: direct Rerun log call, viewer-only proof, stream bypass

Canonical demo camera topics: The first runtime camera visualization demo publishes a single RGB image and camera model on the conventional color_image and camera_info topics. Avoid: demo-only topic namespace, artifact image, direct viewer entity

Damiao-based Robot: A robot whose joints are actuated by one or more Damiao motors, possibly spread across multiple CAN buses and physical limbs. Avoid: Damiao arm when the robot may contain multiple motor groups

Damiao Joint Group: An ordered set of Damiao-driven joints that forms a meaningful physical group such as an arm, torso, or other controllable body section. Avoid: Arm when the group is not necessarily an arm

Damiao Bus: A named communication channel used by a Damiao-based Robot to reach one or more Damiao motors. Avoid: Treating a bus as owned by a single joint group when multiple groups may share a channel

OpenArm: An OpenArm robot configuration built from Damiao motors, with OpenArm-specific joints, side naming, limits, and robot description. Avoid: Damiao robot when referring to OpenArm-specific geometry or naming

Teleop adapter: A device-specific bridge from a human teleoperation source, such as a headset controller, phone, keyboard, or physical leader arm, into DimOS coordinator-facing command streams. Avoid: teleop backend when the component emits coordinator command types, teleop module when referring only to the device adapter, controller when the device is not a robot controller

Teleop profile: A configured teleoperation behavior that selects one primary way human intent drives robot motion while allowing secondary engagement, status, and diagnostic signals. Avoid: backend when referring to behavior selection, mode when the distinction affects routing and safety semantics

Primary motion output: The single motion-control path a teleoperation behavior uses to drive robot movement, so one human input source does not unintentionally command the same robot through multiple motion abstractions at once. Avoid: all active outputs, debug stream, secondary status output

Teleop command envelope: A small wrapper around a coordinator-facing teleoperation command that distinguishes an active command from no command and from an explicit stop command. Avoid: using a missing command as a stop signal, overloading raw motion-message contents with teleop authority state