Progressive Reinforcement Learning with Distillation for Multi-Skilled Motion Control

Authors: Glen Berseth, Cheng Xie, Paul Cernek, Michiel Van de Panne

ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare our progressive learning and integration via distillation (PLAID) method against three alternative baselines.
Researcher Affiliation Academia University of British Colubia
Pseudocode No The paper describes methods and frameworks but does not contain pseudocode or explicitly labeled algorithm blocks.
Open Source Code No The paper does not provide a link or an explicit statement about the availability of its source code.
Open Datasets No The paper describes a simulated environment with a '2D humanoid walker (pdbiped)' and 'randomly generated' terrain types, but it does not use or provide access to a public or open dataset with a concrete link, DOI, or formal citation.
Dataset Splits No The paper does not specify explicit training, validation, or test dataset splits (e.g., percentages or sample counts) to reproduce the data partitioning.
Hardware Specification No The paper states 'Each training simulation takes approximately 5 hours across 8 threads', but it does not specify any particular CPU model, GPU model, or other hardware components used for the experiments.
Software Dependencies No The paper mentions 'Stochastic Gradient Decent (SGD) with momentum' and 'Python', but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes For all of our experiments we linearly anneal ϵ from 0.2 to 0.1 in 100, 000 iterations and leave it from that point on. Each training simulation takes approximately 5 hours across 8 threads. For network training we use Stochastic Gradient Decent (SGD) with momentum. During the distillation step we use gradually anneal the probability of selecting an expert action from 1 to 0 over 10, 000 iterations.