Planning for Learning Object Properties
Authors: Leonardo Lamanna, Luciano Serafini, Mohamadreza Faridghasemnia, Alessandro Saffiotti, Alessandro Saetti, Alfonso Gerevini, Paolo Traverso
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we provide an experimental evaluation in both a simulated and a real environment, which shows that the proposed approach is able to successfully learn how to recognize new object properties. |
| Researcher Affiliation | Academia | Leonardo Lamanna1,2, Luciano Serafini1, Mohamadreza Faridghasemnia3, Alessandro Saffiotti3, Alessandro Saetti2, Alfonso Gerevini2, Paolo Traverso1 1 Fondazione Bruno Kessler, Trento, Italy 2 Department of Information Engineering, University of Brescia, Italy 3 Center for Applied Autonomous Sensor Systems, University of Orebro, Sweden |
| Pseudocode | Yes | Algorithm 1: PLAN AND ACT TO LEARN OBJECT PROPS |
| Open Source Code | No | The paper does not provide a direct link to a source code repository or explicitly state that their implementation code is open-sourced or available. |
| Open Datasets | Yes | We experiment our approach in the ITHOR (Kolve et al. 2017) photo-realistic simulator of four types of indoor environments: kitchens, living-rooms, bedrooms, and bathrooms. |
| Dataset Splits | Yes | For our experiment we split the 120 different environments, provided by ITHOR, into 80 for training, 20 for validation, and 20 for testing. ... The training and validation sets contain 115 object types and are composed by 259859 and 56190 examples, respectively. |
| Hardware Specification | No | The paper mentions using a "Softbank s Pepper humanoid robot" for the real-world demonstrator, but it does not provide specific hardware details (like GPU or CPU models, memory) for the computational resources used to run the main simulation experiments or train models. |
| Software Dependencies | Yes | the neural networks ρt,p s are trained for 10 epochs with 1e 4 learning rate; the other hyperparameters are set to the default values provided by Py Torch1.9 (Paszke et al. 2019). |
| Experiment Setup | Yes | the neural networks ρt,p s are trained for 10 epochs with 1e 4 learning rate; the other hyperparameters are set to the default values provided by Py Torch1.9 (Paszke et al. 2019). We consider that the input object has the property p if the probability is higher than a given threshold (set to 0.5 in our experiments). |