Task-Oriented Active Perception and Planning in Environments with Partially Known Semantics
Authors: Mahsa Ghasemi, Erdem Bulgur, Ufuk Topcu
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the proposed method on various finite-horizon tasks in planar navigation settings where the empirical results show that the proposed method provides reliable task performance that also improves as the knowledge about the environment enhances. and 5. Simulation Results We evaluate the proposed task-oriented active perception and planning algorithm in two simulation domains. |
| Researcher Affiliation | Academia | 1Department of Electrical and Computer Engineering, University of Texas at Austin 2Department of Aerospace Engineering and Engineering Mechanics, University of Texas at Austin. |
| Pseudocode | Yes | Algorithm 1 Constructing an Active Perception Strategy |
| Open Source Code | Yes | Simulation videos, additional simulation results, and link to the implementation files are provided in the supplementary material. |
| Open Datasets | No | In the first set of simulations, we consider an agent that navigates in a discretized 2D environment and has a finite-horizon task. For instance, the task encoded as a DFA in Figure 3 asks the agent to either go to the state where door1 is located or find a key and go to the state where door2 is located, while avoiding the obstacles. We implemented different versions of the task-oriented perception and planning algorithm to evaluate the effect of each module on the performance. Table 1 reports the results for a reach-avoid task in an environment with 64 states and with randomly generated obstacles and target. and In the Air Sim (Shah et al., 2017) simulator, we designed an urban environment and tasked a drone to fly from an initial state to a specific flagged building while avoiding collision with other entities of the environment. No concrete access information for the environment data. |
| Dataset Splits | No | No explicit information about training, validation, or test dataset splits (e.g., percentages, sample counts, or citations to standard splits) is provided. |
| Hardware Specification | No | The paper mentions a drone with '4 cameras and 4 depth sensors' and the 'Air Sim simulator' as part of the simulation environment, but it does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run these simulations or experiments. |
| Software Dependencies | No | The paper mentions 'Air Sim (Shah et al., 2017) simulator' but does not provide specific version numbers for this or any other software dependencies used for the experiments. |
| Experiment Setup | Yes | In the first set of simulations, we consider an agent that navigates in a discretized 2D environment and has a finite-horizon task... an environment with 64 states and with randomly generated obstacles and target. and One of the input parameters to the algorithm is a threshold γd on the above divergence. and Another input parameter to the proposed perception and planning algorithm is a threshold γr on the risk due to perception uncertainty. and This algorithm takes a bound CA on the number of actions and uses that to construct a tree of depth CA. and using a hyperparameter β that weighs safety versus information quality. |