Small starter repo for a vertical-plane tracking task with a fixed-base 3-DoF arm.

• assets/urdf/vertical_arm_3dof.urdf: robot model
• assets/urdf/vertical_arm_4dof.urdf: optional 4-DoF variant
• assets/targets/fixed_targets.csv: end-effector targets in the x-z plane
• scripts/print_robot_info.py: prints joint names and limits
• scripts/view_robot.py: opens the robot with joint sliders
• scripts/visualize_run.py: main runner

Track the provided end-effector targets in the vertical plane using two approaches:

• one classical controller
• one reinforcement learning controller

Assumptions:

• gravity is on
• the task is position-only
• targets are sampled from assets/targets/fixed_targets.csv
• the low-level interface is joint impedance

scripts/visualize_run.py handles the simulation loop and target switching. The controller hook is candidate_impedance_controller().

The controller is expected to return:

• desired joint positions
• desired joint velocities
• kp
• kd

Returning zero gains leaves the arm passive under gravity.

Please submit:

• one classical solution for the tracking task
• one RL-based solution for the tracking task
• a short document summarizing tracking performance
• a comparison between the two approaches

The comparison should include at least:

• tracking accuracy
• stability of motion
• behavior across the provided targets
• any clear failure modes

The write-up does not need to be long, but it should be complete enough to explain:

• how each method was implemented
• how each method was evaluated
• where each method worked well
• where each method struggled

There is also a 4-DoF version of the same arm in assets/urdf/vertical_arm_4dof.urdf.

As a stretch goal, compare the 3-DoF and 4-DoF models for both:

• the classical controller
• the RL controller

The comparison can be qualitative or quantitative, but it should address how the extra degree of freedom changes:

• tracking behavior
• ease of control design
• stability or smoothness
• any differences in RL training behavior

python -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -r requirements.txt

python scripts/print_robot_info.py
python scripts/view_robot.py
python scripts/visualize_run.py

To use the 4-DoF model:

python scripts/print_robot_info.py --robot 4dof
python scripts/view_robot.py --robot 4dof
python scripts/visualize_run.py --robot 4dof

Keep the scope tight. No orientation tracking, obstacle avoidance, or extra infrastructure is needed.