^* denotes equal contribution and ^† denotes equal advising. Representative papers are highlighted.

TL;DR: We observe the value overestimation issue in planner-based MBRL and propose a policy constraint solution with SOTA performance.

TL;DR: ASAP learns agile whole-body humanoid motions via learning a residual action model from the real world to align sim and real physics.

TL;DR: A survey paper covering control, planning, and learning methods for humanoid locomotion and manipulation.

TL;DR: We propose a learning-based locomotion framework that is agile, safe, and adpative in dynamic environments.

TL;DR: DIAL-MPC is the first training-free method achieving real-time whole-body torque control using full-order dynamics.

TL;DR: HOVER unifies many humanoid whole-body control modes to one policy that supports diverse control modes and outperforms each specialist.

TL;DR: AnyCar is a transformer-based dynamics model that can adapt to various vehicles, environments, state estimators, and tasks.

TL;DR: Continuous, agile, and autonomous quadrupedal jumping via hierarchical model-free RL and model-based control.

TL;DR: Pretrain a residual dynamics DNN using meta-learning and fine-tune the whole DNN online using adaptive control with stability guarantees.

TL;DR: Using the distance-based kinematics formulations, we develop an efficient motion planning algorithm PDOMP.

TL;DR: Meta-learn a vehicle dynamics model ensemble and use uncertainty-aware MPPI for control.

TL;DR: We introduce a novel safety reward formulaiton and use Q-value functions to safeguard task policies.

TL;DR: MBD is a diffusion-based traj optimization method that directly computes the score function using models without any external data.

TL;DR: Flying calligrapher enables precise hybrid motion and contact force control for an aerial manipulator in various drawing tasks.

TL;DR: OmniH2O provides a universal whole-body humanoid control interface that enables diverse teleoperation and autonomy methods.

(Oral Presentation)

TL;DR: WoCoCo is a task-agnostic skill learning framework without any motion priors, by decomposing long-horizon tasks into contact sequences.

(Oral Presentation)

TL;DR: H2O enables real-time whole-body teleoperation of a full-sized humanoid to perform tasks like pick and place, walking, kicking, boxing, etc.

(Outstanding Student Paper Award Finalist)

TL;DR: ABS enables fully onboard, agile (>3m/s), and collision-free locomotion for quadrupedal robots in cluttered environments.

TL;DR: We design a terrain-aware MPC framework that enables agile driving over uneven offroad geometries such as hills, banks, and ditches.

TL;DR: We propose a fast and scalable method for high-dim constrained IK problems, based on propagative distance-based optimization.

TL;DR: We quantify the convergence rate of sampling-based MPC, and design a practical and effective algorithm CoVO-MPC with optimal rate.

TL;DR: HMAC handles both manageable and latent disturbances with hierarchical iterative learning and smoothed streaming meta-learning.

TL;DR: SafeDPA jointly tackles the problems of policy adaptation and safe reinforcement learning, under unseen disturbances in the real world.

TL;DR: DMPO learns the inner-loop optimizer of sampling-based MPC directly via experience, outperforming MPC and end-to-end RL baselines.

TL;DR: We introduce a new aerial manipulation system that leverages tactile feedback for accurate contact force control and texture detection.

(Best Paper Award - Innovation)

TL;DR: GYF is a safe falling and recovery framework that can actively tumble and recover to stable modes to reduce damage.

(Spotlight, 3.1%)

TL;DR: Not all model parameters are equally important. We develop an instance-optimal exploration algorithm for MBRL in nonlinear systems.

TL;DR: Inspired by robotics applications, we study algorithms for active representation learning with continuous task parametrization.

(Oral presentation, 6.6%)

TL;DR: DATT can precisely track arbitrary, potentially infeasible trajectories in the presence of large disturbances.

TL;DR: CAJun is a hierarchical learning and control framework that enables legged robots to jump continuously with adaptive distances.

TL;DR: We combine adaptive nonlinear control and neural control for frequency restoration in power systems.

TL;DR: Neural-Fly uses adaptive control to online fine-tune a meta-pretrained DNN representation, enabling rapid adaptation in strong winds.

TL;DR: We propose a new online optimization setting with delay and nonlinear switching cost, and provide compeititve algorithms.

TL;DR: For online control with noisy predictions, we design an algorithm to optimally balance robustness and consistency (performance if no noise).

TL;DR: We study MPC's dynamic regret and competitive ratio in the presence of prediction inaccuracy and feedback delay.

(Spotlight, 2.9%)

TL;DR: We prove MPC's dynamic regret and competitive ratio exponentially improve as its prediction gets longer, in LTV systems.

TL;DR: We present an online multi-task learning approach for adaptive nonlinear control with non-asymptotic guarantees.

TL;DR: We develop a simple and efficient UQ method for 6-DoF pose estimation, and apply it in real-world grasping tasks.

TL;DR: Neural-Swarm is a learning-based controller and planner for close-proximity flight of heterogeneous multirotor swarms.

TL;DR: We derive an iterative algorithm to solve information-cost stochastic nonlinear optimal control problems for safe episodic learning.

TL;DR: We derive generalization bounds under domain shift and connect them with safety bounds in control, for end-to-end safe explorations.

TL;DR: We give the first non-asymptotic guarantee for MPC. MPC's dynamic regret exponentially decreases as its prediction gets longer.

TL;DR: We show competitive algorithms for a new class of online optimization problems, with applications in online competitive control.

TL;DR: We deploy Deep Sets to learn complex interactions between multirotors, for decentralized close-proximity control.

TL;DR: Spectrally normalized deep learning and nonlinear control enable provably stable agile drone landing.

TL;DR: We use inverse RL to learn reward models for human-like autonomous car following.