x_i^{(j)} = [p_i^{j}, v_i^{j}, p_i^{goal,j}, entitytype(j), cos \Delta \theta_{ij}, sin \Delta \theta_{ij}, d_{ij}]

Training

Execution

x_{agg}^{(i)} = GNN(x_i^{i}, x_i^{j} \forall j \in \mathcal{N}(i), g^{(i)})

Aggregation of information from agent's neighborhood using GNN

X_{agg} = \frac{1}{n} \sum_{i=1}^n x_{agg}^{(i)}

proj