Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Agent-Based Modeling for Predicting Pedestrian Trajectories Around an Autonomous Vehicle
Authors: Manon Prédhumeau, Lyuba Mancheva, Julie Dugdale, Anne Spalanzani
JAIR 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We calibrate the model by fitting the parameters values on a training set. We validate the model and evaluate its predictive potential through qualitative and quantitative comparisons with ground truth trajectories. |
| Researcher Affiliation | Academia | Manon Pr edhumeau EMAIL Lyuba Mancheva EMAIL Julie Dugdale EMAIL Univ. Grenoble Alpes, LIG 38000 Grenoble, France Anne Spalanzani EMAIL Univ. Grenoble Alpes, Inria 38000 Grenoble, France |
| Pseudocode | Yes | Appendix C. Algorithms of the Proposed Decision Model Algorithm 2: Main algorithm Constant: radiuscollision = radius AV + radiusp: radius of the collision zone; radiusdanger = radiuscollision + margindanger: radius of the danger zone; radiusrisk = radiuscollision + marginrisk: radius of the risk zone; φ: angle threshold value to distinguish between interaction types; TTCimminent: time threshold value to identify an imminent conflict; TTCdangerconsidered: time threshold value to identify a near-future danger; Input : pos AV : current position of AV ; posp: current position of p; v AV : current velocity of AV ; vp: current velocity of p; vpref p : current preferred velocity of p, i.e. current direction at preferred speed; posgroup: if p is in group, current position of p s group; vgroup: if p is in group, current velocity of p s group; 2 TTCdanger = compute TTCdanger(posp, pos AV , vpref p , v AV , radiusdanger); |
| Open Source Code | No | The model proposed in Section 3 has been implemented in C++ using Pedsim ros (Okal et al., 2014). Pedsim ros is an open source crowd simulator that was adapted to implement the SFM described by Pr edhumeau et al. (2019, 2020), and integrates a set of software libraries and tools to develop robot applications. |
| Open Datasets | Yes | We used the CITR dataset (Yang et al., 2019) as reference data for real pedestrian trajectories... The DUT dataset (Yang et al., 2019) consists of uncontrolled interactions in a crowded campus at Dalian University of Technology, China. |
| Dataset Splits | Yes | We used a 3-block cross-validation method in order to select the best-fitting parameter values so the model can generalize over different datasets... We split the dataset scenarios into a training set (the data known by the model), a test set (data unknown by the model used in the cross-validation loop) and a validation set (data unknown by the model). We randomly selected 12 scenarios for the training and test set, shown in orange in Figure 7 and 4 scenarios for the final validation set, shown in blue in Figure 7. The training and test set was then split in 3 in order to have 8 scenarios for the training set and 4 scenarios for the test set. |
| Hardware Specification | Yes | At 25Hz, a simulation with an AV in a crowd of 100 pedestrians at a density of 0.5 pedestrian/m2 runs in real time on a standard PC hardware (Intel Core i7-7920HQ, 4.10GHz). |
| Software Dependencies | No | The model proposed in Section 3 has been implemented in C++ using Pedsim ros (Okal et al., 2014). |
| Experiment Setup | Yes | The agents were assigned a preferred walking speed following the normal distribution generally observed in real pedestrians with µ = 1.34 m/s and σ = 0.26 m/s (Bosina & Weidmann, 2017)... The set of calibrated parameters {margindanger: 0.45m ; marginrisk: 1.4m ; φ: 25 ; TTCdangerconsidered: [-1,5]s ; TTCimminent: 2s} is very close to the default values fitted by hand (in Table 2). |