A Regret Minimization Approach to Multi-Agent Control

Authors: Udaya Ghai, Udari Madhushani, Naomi Leonard, Elad Hazan

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically evaluate our method on a model of an overactuated aircraft. We show that the distributed method is robust to failure and to adversarial perturbations in the dynamics. Section 5 Experiments
Researcher Affiliation Collaboration 1Department of Computer Science, Princeton University, Princeton, NJ 2Google AI Princeton, Princeton, NJ 3Department of Mechanical and Aerospace Engineering, Princeton, NJ.
Pseudocode Yes Algorithm 1 Multiplayer OCO, Algorithm 2 Multiplayer OCO with Memory, Algorithm 3 Multi-Agent Nonstochastic Control
Open Source Code No Code is adapted from reference GPC implementations in (Gradu et al., 2021). The paper does not provide a direct statement or link for the open-sourcing of its own methodology's code.
Open Datasets No The paper uses the ADMIRE overactuated aircraft model and generates disturbances (Gaussian, random walk, sinusoidal) rather than relying on a pre-existing public dataset with concrete access information.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) for training, validation, or testing.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions the use of GPC, LQR, Hinf control, and MAGPC algorithms, and states 'Code is adapted from reference GPC implementations in (Gradu et al., 2021),' but it does not provide specific version numbers for any software, libraries, or frameworks used.
Experiment Setup Yes GPC has a decaying learning rate of 0.001/t, a rollout length h of 5 and uses DACs with windows of size 5. MAGPC is split into 4 1-d controllers, each using decaying learning rate of 0.001/t, a rollout length 5 and a DAC with window length 5. Disturbance details: 1. Random walk chooses wt = wt-1 + Xt where Xt is a standard Gaussian random variable. 2. Gaussian noise is iid. and has variance 1. 3. Sinusoidal noise is chosen via wi = sin(2t + φi) where φ = [12.0, 21.0, 3.0, 42.0, 1.0], picked arbitrarily once. Fourth control is set to 0 in one experiment and otherwise controls are left unchanged.