The Value of Sensory Information to a Robot
Authors: Arjun Krishna, Edward Hu, Dinesh Jayaraman
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Towards this, we empirically study agents of varying architectures, generated with varying policy synthesis approaches (imitation, RL, model-based control), on diverse robotics tasks. For each robotic agent, we characterize its regret in terms of performance degradation when state observations are withheld from it at various task states for varying lengths of time. |
| Researcher Affiliation | Academia | Arjun Krishna University of Pennsylvania EMAIL Edward S Hu University of Pennsylvania Dinesh Jayaraman University of Pennsylvania |
| Pseudocode | Yes | The pseudocode for the proposed greedy strategy ηgreedy is provided in Algorithm 1. |
| Open Source Code | No | The code used to conduct the experiments will be made available on the website. |
| Open Datasets | Yes | We study 7 diverse robotic tasks, depicted in Figure 2, sourced from DM-control (Tassa et al., 2018), Robosuite (Zhu et al., 2020; Mandlekar et al., 2023), and the Push-T task (adopted from Chi et al. (2023)). |
| Dataset Splits | No | The paper uses established benchmark tasks and describes data collection methods (e.g., 200 demos for Robosuite, 200 trajectories from closed loop execution), but it does not specify explicit training, validation, and test splits with percentages or sample counts for the models being trained. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models, memory, or cloud computing instance types used for running the experiments. It mentions 'on-robot execution' but this is too general. |
| Software Dependencies | No | The paper mentions software like 'JAX implementation of the TD-MPC2 algorithm' and 'code adopted from the official repository (Chi et al., 2023) for training U-Net based DP agents', but it does not provide specific version numbers for JAX, Python, or any other libraries or frameworks. |
| Experiment Setup | Yes | Table 2: Default hyperparameters of TD-MPC2 agent [...values for encoder.layers, encoder.lr, world model.mlp-dim, etc...] Table 5: Default hyperparameters of Diffusion Policy for a task of episode length T [...values for input embed dim, num diffusion steps, batch size, lr, etc...] |