Embedded Files
Mobile Pedipulation for Object Sliding via Hierarchical Control on a Wheeled Bipedal Robot
We introduce a hierarchical control framework that enables wheeled bipedal robots to slide objects with their legs, combining robot-object motion planning, nonlinear model predictive control based on a a novel reduced-order model, and whole-body control. Hardware experiments demonstrate under-desk object retrieval and dynamic scooting.
Robotics and Automation Letter (RA-L) 2026 | Website | arXiv | Video |
SurGE: Surrogate Gradient-guided Evolution for Co-design of Legged Robots with Parallel Elasticity
SurGE is a co-design framework that guides an evolutionary search with an gradient from a differentiable model and a design aware control policy to jointly optimize the spring hardware and the controller, reducing the design objective by 37.65% on hardware while converging about 6 times more consistently than standard evolutionary search.
This work introduces a kinodynamic model predictive control (MPC) framework that exploits unidirectional parallel springs (UPS) to improve the energy efficiency of dynamic legged robots. Preliminary hardware experiments show a 14.8% reduction in energy consumption.
Contact Sensing via Joint Torque Sensors and a Force/Torque Sensor for Legged Robots
We propose a generalized momentum-based observer framework for detecting and localizing contact with data from (a) a hip-mounted force-torque (FT) sensor and (b) distributed low-cost strain-gauge-based joint torque sensors.
CASE 2025
Design and development of the MIT humanoid: A dynamic and robust research platform
This paper introduces the MIT Humanoid, a 1 m tall, 24 kg research platform with 18 actuated degrees of freedom driven by custom high torque proprioceptive motor modules, designed for the power density and mechanical robustness that parkour style dynamic motions demand. Initial hardware experiments with model based controllers demonstrate stable pose control, walking, and vertical jumps of roughly 30 cm.
Humanoids 2023 | Video |
Orientation-aware model predictive control with footstep adaptation for dynamic humanoid walking
This paper introduces a new control method that lets a humanoid robot decide where to place its feet and how to swing its torso at the same time, instead of treating those choices separately. In simulation, the approach keeps the MIT Humanoid steadier when it gets pushed or twisted and lets it walk across uneven, wavy ground.
Humanoid Self-Collision Avoidance Using Whole-Body Control with Control Barrier Functions
The paper combines control barrier functions (CBFs) with a whole-body controller (WBC) so the MIT Humanoid can guarantee collision-free, joint-limit-respecting motion in real time by enforcing constraints via CBF inequality constraints in the quadratic program (QP).
Representation-Free Model Predictive Control for Dynamic Motions in Quadrupeds
This article presents a representation-free model predictive control (RF-MPC) framework for controlling various dynamic motions of a quadrupedal robot in three-dimensional (3-D) space, where rotation matrix is directly used for rotational dynamics.
Best Paper Award Finalist from Technical Committee (TC) on Optimization for Robotics
Transactions on Robotics (T-RO) | Video | Code | Preprint |
ICRA 2019 | Video |
Hoppy: An open-source kit for education with dynamic legged robots
This paper presents HOPPY, an open source, low cost, and modular kit in which a single leg hops around a rotating gantry, intended to lower the barrier to studying legged locomotion on real hardware by connecting the theory taught in fundamental robotics courses to software and hardware integration. A heuristic controller achieves speeds up to 1.7 m/s while traversing small obstacles and rejecting disturbances with a counterweight, and the kit was used for a semester long project in an undergraduate robot dynamics and control course.
Hybrid Sampling/Optimization-based Planning for Agile Jumping Robots on Challenging Terrains
We propose a two-stage kinodynamic jumping motion planner: the sampling stage proposes candidate jump sequences guided by a reachability map, while a trajectory optimization stage generates dynamically feasible motions.
ICRA 2021 | Video | Preprint |
Kinodynamic Motion Planning for Multi-Legged Robot Jumping via Mixed-Integer Convex Program
IROS 2020 | Video |
Centroidal-momentum-based trajectory generation for legged locomotion
Mechatronics 2020 |
Single Leg Dynamic Motion Planning with Mixed-Integer Convex Optimization
IROS 2018 | Video |
Design and Experimental Implementation of a Quasi-Direct-Drive Leg for Optimized Jumping
This paper presents an actuator design method for jumping robots that couples control design for optimal ground reaction force with mechanical selection of motor and gearbox pairs in a single nonlinear optimization, so that the torque and speed capacity of electromagnetic motors is fully exploited. A two degree of freedom leg prototype built from the resulting actuators reached a vertical jump height of 0.62 m and a forward jump distance of 0.72 m, corresponding to 2.4 and 2.7 times its leg length.
IROS 2017 | Video |
Page updated
Report abuse