Deep Imitation Learning for Bimanual Robotic Manipulation

Authors: Fan Xie, Alexander Chowdhury, M. Clara De Paolis Kaluza, Linfeng Zhao, Lawson Wong, Rose Yu

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate on two variations of a table-lifting task where bimanual manipulation is essential. We show that both hierarchical and relational modeling provide significant improvement for task competition on held-out instances. For the task shown in Figure 1, our model achieves 29% success rate, whereas a baseline approach without these contributions only succeeds in 1% of the test cases.
Researcher Affiliation Academia 1Northeastern University, Boston MA, USA, 2University of California, San Diego, USA.
Pseudocode No The paper describes various models and algorithms but does not include any structured pseudocode or algorithm blocks (e.g., a figure or section explicitly labeled 'Pseudocode' or 'Algorithm').
Open Source Code Yes We open source the code for simulation, data, and models at: https://github.com/Rose-STL-Lab/HDR-IL.
Open Datasets No The paper describes generating its own simulation data from manually labeled demonstrations ('Demonstrations were manually labeled as separate primitives in the simulator and sequenced together to generate the simulation data.') and training on it ('We generate a total of 2,500 training demonstration trajectories'), but does not provide concrete access information (link, DOI, formal citation) for a publicly available or open dataset.
Dataset Splits No The paper states 'We generate a total of 2,500 training demonstration trajectories' and 'We tested on a random sample of 127 starting points', but does not explicitly mention a separate validation set or specific split percentages for training, validation, and testing.
Hardware Specification No The paper mentions conducting simulations 'using the Py Bullet [42] physics simulator with a Baxter robot' but does not provide specific details on the hardware (e.g., GPU/CPU models, memory) used to run these simulations.
Software Dependencies No The paper mentions 'Py Bullet [42] physics simulator' but does not provide specific version numbers for this or any other software dependencies needed to replicate the experiment.
Experiment Setup Yes All models were trained with the ADAM optimizer with batch size 70 over 12,500 epochs.