PELE Musculoskeletal Leg Control

End-to-end electrohydraulic leg system integration with 500 Hz real-time cascaded task-space control.

Outcome

I led the control implementation for PELE, a bio-inspired electrohydraulic musculoskeletal leg designed for agile, adaptive, and energy-efficient locomotion. I developed and validated a real-time C++ cascaded controller that maps task-space goals to joint, tendon, and actuator-level commands on hardware.

Problem

Conventional legged robots often rely on rigid electromagnetic actuation with heavy sensing and control stacks to handle uneven terrain. PELE instead uses compliant electrohydraulic muscles, which introduced a control challenge: maintain stable, accurate motion while handling nonlinear tendon-driven dynamics and high-voltage actuator interfaces in real time.

System

  • Mechanics: carbon-fiber leg with hip/knee joints, tendon routing, and antagonistic electrohydraulic muscle packs
  • Electronics: computer + DAQ + high-voltage amplifiers driving four muscle packs (hip/knee flexor-extensor pairs)
  • Sensing: joint encoders plus capacitive self-sensing from voltage/current measurements
  • Control: cascaded controller from task-space planning to joint control to low-level actuator voltage commands
  • Runtime software: multithreaded C++ pipeline for DAQ communication, control loop, filtering, and logging

The cascaded control structure is shown below for context.

Contribution

  • Controller architecture and implementation in C++
  • Integration of sensing, DAQ, and high-voltage actuator drive with the control stack on the existing mechanical platform
  • Experimental validation and performance analysis

Technical Stack

  • Real-time control system in C++ (500 Hz loop)
  • Cascaded task-space, joint-space, and actuator-level control
  • Musculoskeletal robotic platform integration
  • High-voltage actuator drive integration
  • DAQ and multithreaded runtime pipeline

Key Results

  • Cascaded closed-loop controller (task space -> joint space -> tendon space/PID) implemented in multithreaded C++ at 500 Hz
  • Closed-loop leg-tip tracking of four predefined trajectories (ellipse, rectangle, infinity, star)
  • Dynamic experiments reached 128 mm jump height (~40% of leg height), with vertical jumping agility of 0.75 m/s and jump frequency of 5.8 Hz
  • In periodic feed-forward force-control mode, the platform achieved 5 Hz gait motion and 10 Hz linear motion
  • Published in Nature Communications (2024)

Media

Supplementary Movie 1: overview of the PELE platform and locomotion behavior.

Impact and Future Direction

This work is a step toward untethered artificial-muscle robots that can move through unstructured environments such as natural terrain and disaster-relevant settings. The current boom-mounted platform validates the control and actuation principles at system level; ongoing work targets integration into full bipedal or quadrupedal robots for deployment beyond lab setups. The same architecture is promising for long-duration operation because it combines agility with low energy demand.

Skills

C++ real-time control task-space control hardware-software integration high-voltage electronics DAQ experiments