> Use whenever a MATLAB robot model (rigidBodyTree) is needed — whether for simulation, visualization, IK, motion planning, pick-and-place, or trajectory UR10, UR10e, UR16e, UR20, KUKA iiwa, Fanuc, ABB, Panda, Kinova, Sawyer, Baxter), robot modeling verbs (load, create, build, import, simulate, model), tasks implying a robot model (pick, place, lift, reach, move, grasp, plan motion for, animate), kinematics keywords (IK, FK, inverse kinematics, forward kinematics, joint configuration, end-effector pose, gripper), or working with rigidBodyTree, loadrobot, importrobot, URDF, DH parameters, addVisual, addCollision.
npx skills add https://github.com/matlab/matlab-agentic-toolkit --skill matlab-model-robot-kinematics
Build manipulator models correctly and validate every kinematic solution.
inverseKinematics vs generalizedInverseKinematicscontopptraj, cubicpolytraj, trapveltraj) — use a trajectory skillreferences/parallel-robot-guidance.mdinverseDynamics, forwardDynamics) — separate domainChoose the highest-fidelity source available:
| Priority | Source | When to Use | Why |
|----------|--------|-------------|-----|
| 1 | loadrobot | Robot exists in the library | Includes collision meshes, inertias, visuals |
| 2 | importrobot | User has URDF/Xacro/SDF file | Preserves mesh references |
| 3 | Build from DH | Only DH parameters available | No collision meshes unless added manually |
Critical: A robot without collision meshes will cause downstream motion planning to silently report "collision-free" paths that actually collide. Always prefer sources that include meshes.
If the user says "build a robot" but names a known robot (UR5e, KUKA iiwa, Panda, etc.), suggest loadrobot first.
After importing from URDF, Simscape, or CAD: Check if collision meshes are present. Simscape Multibody models and many URDFs/CAD exports only provide visual geometry — no collision meshes. If collision meshes are missing, read references/import-mesh-handling.md for the decision tree on generating them from visuals.
Always set DataFormat="row" — this is required for planning and dynamics workflows.
Option A — loadrobot:
robot = loadrobot("universalUR5e", DataFormat="row");
Option B — importrobot (URDF/Xacro/SDF):
robot = importrobot("myRobot.urdf", DataFormat="row");
If the URDF has only visual geometry (no <collision> tags), add VHACD collision decomposition at import time:
opts = vhacdOptions("RigidBodyTree");
opts.SourceMesh = "VisualGeometry";
robot = importrobot("myRobot.urdf", DataFormat="row", ...
MeshPath="path/to/meshes", CollisionDecomposition=opts);
Option C — Build from DH parameters:
robot = rigidBodyTree(DataFormat="row");
body = rigidBody("link1");
jnt = rigidBodyJoint("joint1", "revolute");
setFixedTransform(jnt, [a alpha d 0], "dh");
jnt.PositionLimits = [-pi, pi];
body.Joint = jnt;
addVisual(body, "Cylinder", [0.04, 0.3]);
addCollision(body, "Cylinder", [0.05, 0.3]);
addBody(robot, body, "base");
tool0 Exists with Z = OutwardBefore attaching a gripper, the robot MUST have a tool0 body whose local Z-axis points along the arm's outward direction (away from the wrist). Most loadrobot models satisfy this already — but KUKA iiwa models do not.
Check references/ee-frame-alignment.md for the full compatibility table and diagnostic procedure.
If the robot already has tool0 with Z = outward (all UR, ABB, Techman, FANUC, Kinova, Yaskawa): proceed directly to gripper attachment.
If the robot lacks tool0 or EE Z is misaligned (KUKA iiwa 7/14): add a corrective tool0 frame:
% KUKA iiwa: iiwa_link_ee Z is perpendicular to arm — fix with Ry(90°)
tool0 = rigidBody("tool0");
tool0Joint = rigidBodyJoint("tool0_joint", "fixed");
R_correction = [0 0 1; 0 1 0; -1 0 0]; % Ry(90°)
setFixedTransform(tool0Joint, rotm2tform(R_correction));
tool0.Joint = tool0Joint;
addBody(robot, tool0, "iiwa_link_ee");
For unknown robots (URDF imports): Run the diagnostic check from references/ee-frame-alignment.md to verify alignment before attaching anything.
Visuals vs collisions — both are needed, for different purposes:
| Function | Purpose | Used By |
|----------|---------|---------|
| addVisual(body, shape, dims) | Display appearance | show for visualization |
| addCollision(body, shape, dims) | Planning geometry | checkCollision, motion planners |
To color a visual: addVisual(body, "Mesh", stlFile, tform, FaceColor=[1 0.8 0])
Robots from loadrobot already have both. When building from scratch, add both explicitly.
Attach a gripper (MANDATORY for manipulation tasks):
If the task involves object interaction (picking, placing, lifting, grasping, manipulating), a gripper is required — do not skip this step. Only omit for pure visualization or reachability studies with no object contact.
Before building ANY custom gripper, you MUST read references/gripper-models.md. Available models include both parallel-jaw (robotiq2F85, 85mm opening) AND vacuum grippers (robotiqEPick variants for large/flat objects). Do NOT assume robotiq2F85 is the only option.
references/gripper-models.mdgripper = loadrobot("robotiq2F85", DataFormat="row");
addSubtree(robot, "tool0", gripper, ReplaceBase=false);
Critical: Always use ReplaceBase=false. The default (true) merges the gripper's base link into the parent body, silently discarding the base visual mesh (the gripper housing/coupling). All provided grippers have a base visual that is lost without this flag.
After attaching a gripper, add a contact frame at the gripper's contact point (fixed joint, zero DOF added), then solve IK to that frame. This avoids confusing tip frame orientations and manual offset math. See references/gripper-models.md for per-gripper offsets and the full pattern.
contactFrame = rigidBody("contact_point");
contactJoint = rigidBodyJoint("contact_joint", "fixed");
setFixedTransform(contactJoint, trvec2tform([0, 0, gripperOffset]));
contactFrame.Joint = contactJoint;
addBody(robot, contactFrame, "tool0");
Add a custom end-effector frame (when no gripper is attached or you need a specific offset):
ee = rigidBody("tool_tip");
eeJoint = rigidBodyJoint("tool_tip_joint", "fixed");
setFixedTransform(eeJoint, trvec2tform([0.1, 0, 0]));
ee.Joint = eeJoint;
addBody(robot, ee, "tool0");
Use FK and IK to confirm the model is built correctly — correct link lengths, joint axes, limits, and end-effector frame placement.
Verify with FK: Move to a known configuration and check the end-effector reaches the expected position.
q = homeConfiguration(robot);
tform = getTransform(robot, q, "tool0");
fprintf("Home position: [%.3f, %.3f, %.3f] m\n", tform(1:3,4));
Visually confirm the pose makes sense:
figure;
show(robot, q, Frames="off");
axis auto;
title("Home configuration");
Verify with IK: Pick a target you know is reachable and confirm the solver converges.
Decision — IK vs GIK:
| Use | When |
|-----|------|
| inverseKinematics | Single target pose, no extra constraints |
| generalizedInverseKinematics | Multiple constraints (position + aiming, joint bounds, orientation, etc.) |
See references/gik-constraints.md for all available GIK constraint types.
ik = inverseKinematics(RigidBodyTree=robot);
weights = [0.25 0.25 0.25 1 1 1];
targetPose = trvec2tform([0.4, 0.1, 0.3]) * eul2tform([0 pi 0], "ZYX");
[qSol, solnInfo] = ik("tool0", targetPose, weights, homeConfiguration(robot));
Never skip this step. Agents consistently skip validation and deliver solutions that silently failed.
After inverseKinematics:
if solnInfo.ExitFlag <= 0
warning("IK did not converge. ExitFlag: %d, Status: %s", ...
solnInfo.ExitFlag, solnInfo.Status);
end
tformActual = getTransform(robot, qSol, "tool0");
posError = norm(tformActual(1:3,4)' - targetPose(1:3,4)');
if posError > 1e-3
warning("IK position error %.4f m exceeds threshold.", posError);
end
After generalizedInverseKinematics:
if solInfo.ExitFlag <= 0
warning("GIK did not converge. ExitFlag: %d, Status: %s", ...
solInfo.ExitFlag, solInfo.Status);
end
for i = 1:numel(solInfo.ConstraintViolations)
cv = solInfo.ConstraintViolations(i);
if cv.Violation > 1e-3
warning("Constraint %d (%s) violated: %.4f", i, cv.Type, cv.Violation);
end
end
For sequential waypoints — check joint continuity:
for i = 2:size(qAll, 1)
maxJump = max(abs(qAll(i,:) - qAll(i-1,:)));
if maxJump > deg2rad(30)
warning("Joint jump of %.1f deg between waypoints %d and %d.", ...
rad2deg(maxJump), i-1, i);
end
end
| Function | Purpose | Toolbox | Available From |
|----------|---------|---------|----------------|
| loadrobot | Load built-in robot with meshes | Robotics System Toolbox | R2019b |
| importrobot | Import from URDF/Xacro/SDF/Simscape | Robotics System Toolbox | R2017a |
| rigidBodyTree | Create empty robot model | Robotics System Toolbox | R2016b |
| setFixedTransform | Set joint transform (DH or homogeneous) | Robotics System Toolbox | R2016b |
| addVisual | Add visual geometry for display | Robotics System Toolbox | R2019a |
| addCollision | Add collision geometry for planning | Robotics System Toolbox | R2019b |
| addSubtree | Attach subtree (gripper) to body | Robotics System Toolbox | R2016b |
| getTransform | Compute FK for a configuration | Robotics System Toolbox | R2016b |
| inverseKinematics | Solve IK for single target pose | Robotics System Toolbox | R2016b |
| generalizedInverseKinematics | Solve IK with multiple constraints | Robotics System Toolbox | R2017a |
| show | Visualize robot configuration | Robotics System Toolbox | R2016b |
figure;
show(robot, q, Frames="off");
axis auto;
view(45, 30);
title("Robot at configuration q");
For multiple configurations, use tiledlayout/nexttile:
figure;
tiledlayout(1, 3);
for i = 1:3
nexttile;
show(robot, qAll(i,:), Frames="off", PreservePlot=false);
axis auto;
title(sprintf("Waypoint %d", i));
end
When solving IK for sequential waypoints, use the previous solution as the initial guess:
qPrev = homeConfiguration(robot);
for i = 1:numWaypoints
[qSol, solnInfo] = ik("tool0", targetPoses(:,:,i), weights, qPrev);
% ... validate ...
qPrev = qSol;
end
DataFormat="row" on creation or importdeg2rad when specifying in degrees)Frames="off" and axis auto for clean visualizationtiledlayout/nexttile instead of subplot"tool_tip", "bucket_tip")inverseKinematics: [orientation(3) position(3)] — set position weights higher for position-priority tasks| Mistake | Why It's Wrong | Correct Approach |
|---------|---------------|-----------------|
| Building from DH when loadrobot has the robot | No collision meshes → planning silently fails | Check loadrobot library first |
| Using addCollision for visual appearance | Collision geometry is simplified, not rendered by show | Use addVisual for display, addCollision for planning |
| Skipping IK validation | Solution may not converge; ExitFlag <= 0 means failure | Always check ExitFlag and verify position error via FK |
| Using DataFormat="column" or "struct" | Incompatible with planning/trajectory functions that expect row vectors | Always use DataFormat="row" |
| Skipping gripper for manipulation tasks | Object interaction (pick, lift, grasp) requires a gripper on the model | Always attach a gripper when the task involves object contact |
| Solving IK to "tool0" or tip body without a contact frame | "tool0" is the flange (offset from contact point); tip body frames have confusing 90° rotations | Add a fixed "contact_point" frame at the gripper offset, solve IK to that |
| Building a custom gripper from scratch | Provided models have accurate geometry and are ready to attach | Check references/gripper-models.md for available grippers |
| Using robotiq2F85 for objects > 85 mm | Jaw cannot open wide enough to grasp the object | Compare object dimensions to gripper max opening; use vacuum for large objects |
| Using IK when multiple constraints are needed | inverseKinematics only handles a single pose target | Use generalizedInverseKinematics with constraint objects |
| Not checking joint continuity for waypoint sequences | Large jumps between solutions cause unsafe trajectories | Compare consecutive solutions, flag jumps > 30 deg |
| Using addSubtree without ReplaceBase=false | Gripper base visual mesh is silently discarded; housing disappears from visualization | Always pass ReplaceBase=false when attaching grippers |
| Attaching gripper to KUKA iiwa without frame correction | iiwa_link_ee Z is perpendicular to arm; gripper points sideways | Add tool0 with Ry(90°) correction before attaching — see references/ee-frame-alignment.md |
| Assuming all loadrobot models have tool0 with Z = outward | Some robots (KUKA iiwa) lack tool0 or have misaligned EE frames | Check references/ee-frame-alignment.md table or run diagnostic before gripper attachment |
| Using rigidBodyTree for closed-loop parallel robots | Cannot represent closed kinematic chains | See references/parallel-robot-guidance.md |
----
Copyright 2026 The MathWorks, Inc.
----
Expert startup business analyst specializing in market sizing, financial modeling, competitive analysis, and strategic planning for early-stage companies. Use PROACTIVELY when the user asks about market opportunity, TAM/SAM/SOM, financial projections, unit economics, competitive landscape, team planning, startup metrics, or business strategy for pre-seed through Series A startups.
This skill should be used when the user asks to "plan team structure", "determine hiring needs", "design org chart", "calculate compensation", "plan equity allocation", or requests organizational design and headcount planning for a startup.
End-to-end bulk RNA-seq orchestrator — takes raw FASTQ reads through QC and trimming (FastQC, fastp/Trim Galore), alignment and quantification (STAR, Salmon, featureCounts), assembles a gene-level counts matrix, then hands off to differential expression (pydeseq2), pathway/GSEA enrichment (pathway-enrichment), and publication figures (scientific-visualization). Use whenever the user has bulk RNA-seq reads or quant output and wants a complete, reproducible differential-expression workflow — e.g. "analyze my RNA-seq", "FASTQ to DESeq2", "run nf-core/rnaseq", "STAR/Salmon quantification", "build a counts matrix for DESeq2", or "go from reads to differentially expressed genes and enriched pathways". Routes between an nf-core/rnaseq (Nextflow) path and a standalone STAR/Salmon path, and covers experimental design, strandedness, and QC gates. For single-cell RNA-seq use the scanpy skill instead.
Generate project status reports from Jira issues and publish to Confluence. When an agent needs to: (1) Create a status report for a project, (2) Summarize project progress or updates, (3) Generate weekly/daily reports from Jira, (4) Publish status summaries to Confluence, or (5) Analyze project blockers and completion. Queries Jira issues, categorizes by status/priority, and creates formatted reports for delivery managers and executives.
Evaluates market bubble risk through quantitative data-driven analysis using the revised Minsky/Kindleberger framework v2.1. Prioritizes objective metrics (Put/Call, VIX, margin debt, breadth, IPO data) over subjective impressions. Features strict qualitative adjustment criteria with confirmation bias prevention. Supports practical investment decisions with mandatory data collection and mechanical scoring. Use when user asks about bubble risk, valuation concerns, or profit-taking timing.
Google Workflow: Today's meetings + open tasks as a standup summary.
Read event data from a Google Sheets spreadsheet and create Google Calendar entries for each row.
Create professional, dark-themed SVG diagrams of any type — architecture diagrams, flowcharts, sequence diagrams, structural diagrams, mind maps, timelines, illustrative/conceptual diagrams, and more. Use this skill whenever the user asks for any kind of technical or conceptual diagram, visualization of a system, process flow, data flow, component relationship, network topology, decision tree, org chart, state machine, or any visual representation of structure/logic/process. Also trigger when the user says "画个图" "画一个架构图" "diagram" "flowchart" "sequence diagram" "draw me a ..." or uploads content and asks to visualize it. Output is always a standalone .svg file.
Take matlab/matlab-model-robot-kinematics from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.