Challenges in Dynamic Legged Locomotion


Organizers: Diego Pardo, Michael Posa, Scott Kuindersma

Website: https://leggedrobotlocomotion.wordpress.com/

The performance of modern legged robots still pales in comparison to their biological counterparts in terms of speed, robustness, versatility, and efficiency. The technical challenges that fuel this gap touch on topics ranging from actuator design to contact modeling and hybrid optimal control. This workshop aims to bring together researchers working on a span of topics at the frontier of legged robotics research relating to design, modeling, estimation, and control.

The dynamic properties of legged locomotion (i.e. nonlinear, underactuated, and hybrid) pose significant theoretical, computational, and physical challenges. Traditional approaches to tackling this dynamic complexity have used a combination of model simplification and factoring of planning, estimation, and control into a sequence of computationally-expedient sub-problems. By ignoring aspects of the robot dynamics and/or removing the ability to reason about feedback during planning, these simplifications degrade performance. However, we still lack a detailed understanding as to what dynamic complexity/computation trade-off is necessary to achieve a given behavior specification, and how these choices are affected by robot design.

The workshop will be framed by the following questions and technical topics:

  • What is the impact of model simplifications and assumptions on the efficiency, robustness and performance of the solutions?
  • What are the key physical limitations of modern legged robots?
  • Should we be designing control systems and hardware together?
  • Are we overlooking the complexity of the contact model or are current approximations sufficient?
  • Topics:
    • Trajectory optimization for legged locomotion
    • Modeling frictional contact for planning, feedback control, and estimation
    • Planning in reduced spaces (e.g., centroidal dynamics, SLIP)
    • Feedback motion planning; optimizing robustness
    • Model-predictive control (e.g., DDP variants)
    • Verification/analysis of motions involving rigid contact
    • Actuator and mechanism design for legged robots
    • Whole-body state estimation