โ† All projects
Robotics ยท Force control

Multi-Step Robotic Arm Control

Peg-in-hole, obstacle avoidance & force-regulated manipulation

The ME 446 final synthesizes a full semester of manipulator control into one run: the arm must move in straight lines, insert a peg, thread a zig-zag obstacle course, and press an egg hard enough to register but not to crack it โ€” all judged on time and trajectory quality.

Role
Robotics / controls
Course
ECE 489 / ME 446
Timeframe
Jan โ€“ May 2024
Platform
TMS320F28335 DSP
Robot arm performing the task sequence
// The arm mid-sequence, controlled entirely in task space.

01 Overview

Everything runs in task space: I derived and verified forward and inverse kinematics in MATLAB, calibrated joint encoder offsets, and implemented the control loop in C on a TMS320F28335 DSP via Code Composer Studio, updating every 1 ms (1 kHz). A Simulink interface gave live visualization and on-the-fly gain tuning. The result is a single state machine that sequences five very different sub-tasks, each demanding a different control strategy.

02 The task sequence

Straight-line moves

Cartesian point-to-point motion via task-space PD control, with the Jacobian transpose mapping desired Cartesian forces into joint torques.

Peg insertion

Insert the peg to a marked depth and hold 0.5 s. Z-axis force regulation with deliberately softened gains (impedance tuning) lets it seat smoothly without fighting the hole.

Zig-zag obstacle course

An inverse-dynamics inner loop cancels the arm's nonlinearities; an outer PD + feedforward loop tracks the path. Gains are weakened perpendicular to the zig-zag axis using frame transformations โ€” impedance control in a rotated frame โ€” so the peg stays compliant where it needs to bend.

Egg pressing (500โ€“1000 g)

Force-based impedance control in Z, with the commanded vertical force applied through the Jacobian transpose and friction compensation tuned to stop the arm from "bouncing" on contact.

Return home

A damped task-space trajectory removes contact force and returns to the calibrated rest pose without jerk or overshoot.

03 Control architecture

  • Joint-space PD + feedforward โ€” baseline trajectory tracking.
  • Friction compensation โ€” applied to every torque command for low jitter.
  • Inverse-dynamics control โ€” cancels nonlinear coupling during the zig-zag.
  • Task-space PD โ€” peg insertion and obstacle clearance.
  • Impedance control โ€” compliant behavior in rotated frames + egg press.
  • Z-force via Jacobian transpose โ€” precise contact force on the egg.

04 Results

ยฑ50 g
Force-mapping accuracy
1 kHz
Real-time control loop
5
Distinct control regimes

Achieved sub-centimeter peg alignment and insertion-depth control, navigated the zig-zag with axis-specific impedance shaping, and held egg force within a ยฑ50 g margin โ€” a robust hybrid of model-based and feedback control with low joint-torque jitter thanks to friction compensation.