Research fields

Robotics

Robotics labs go from an assembly to motion and learned control without leaving LabSpace. Work out kinematics and controllers in notebooks, analyse logged runs, run assemblies through their joints and train reinforcement-learning policies on them, with the compute cost in view.

LabSpace for robotics labs

• Solve a two-link arm's kinematics

  Write the forward kinematics of a two-link arm, solve the joint angles for a target, and check that the arm can follow a path before you program it.

  1. Write the arm's forward kinematics
  2. Solve the joint angles for a target
  3. Track a circular path
  4. Plot the arm along the path

• Tune a joint controller

  Simulate a joint's step response under different controller gains in a notebook, and compare overshoot and settling time before you change the firmware.

  1. Model the joint as a second-order system
  2. Simulate step responses for three gain settings
  3. Measure overshoot and settling time
  4. Plot the responses side by side

• Analyse logged trajectories

  Load joint logs from a run and compare the commanded and measured paths, with the tracking error for each joint.

  1. Load the run's joint log
  2. Compare commanded and measured angles
  3. Compute tracking error per joint
  4. Plot the worst joint over time

The tools behind it

• Motion simulation: Assemblies run through their joints to check motion before a build
• Train a policy: Reinforcement learning on your assembly, packaged when training finishes
• Compute: Your own cloud account with quotas and a monthly cost ceiling
• Notebooks: Script the CAD model and analyse runs from Python