{ "cells": [ { "cell_type": "code", "execution_count": 24, "metadata": {}, "outputs": [], "source": [ "import torch\n", "import dataclasses\n", "import mediapy\n", "from huggingface_hub import PyTorchModelHubMixin\n", "from huggingface_hub import ModelCard\n", "from gpudrive.networks.late_fusion import NeuralNet\n", "\n", "from gpudrive.env.config import EnvConfig\n", "from gpudrive.env.env_torch import GPUDriveTorchEnv\n", "from gpudrive.visualize.utils import img_from_fig\n", "from gpudrive.env.dataset import SceneDataLoader\n", "from gpudrive.utils.config import load_config" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Configs" ] }, { "cell_type": "code", "execution_count": 25, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "{'max_controlled_agents': 64, 'ego_state': True, 'road_map_obs': True, 'partner_obs': True, 'norm_obs': True, 'remove_non_vehicles': True, 'lidar_obs': False, 'reward_type': 'weighted_combination', 'collision_weight': -0.75, 'off_road_weight': -0.75, 'goal_achieved_weight': 1.0, 'dynamics_model': 'classic', 'collision_behavior': 'ignore', 'dist_to_goal_threshold': 2.0, 'polyline_reduction_threshold': 0.1, 'sampling_seed': 42, 'obs_radius': 50.0, 'action_space_steer_disc': 13, 'action_space_accel_disc': 7, 'init_mode': 'all_non_trivial'}\n" ] } ], "source": [ "# Configs model has been trained with\n", "config = load_config(\"../../examples/experimental/config/reliable_agents_params\")\n", "\n", "print(config)\n", "\n", "max_agents = config.max_controlled_agents\n", "num_envs = 2\n", "device = \"cpu\" # cpu just because we're in a notebook" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Load pre-trained agent via Hugging Face hub\n" ] }, { "cell_type": "code", "execution_count": 26, "metadata": {}, "outputs": [], "source": [ "sim_agent = NeuralNet.from_pretrained(\"daphne-cornelisse/policy_S10_000_02_27\")" ] }, { "cell_type": "code", "execution_count": 27, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "91" ] }, "execution_count": 27, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# Agent has an action dimension of 91: 13 steering wheel angle discretizations x 9 acceleration discretizations\n", "sim_agent.action_dim" ] }, { "cell_type": "code", "execution_count": 28, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "2984" ] }, "execution_count": 28, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# Size of flattened observation vector\n", "sim_agent.obs_dim" ] }, { "cell_type": "code", "execution_count": 29, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "['ffn', 'model_hub_mixin', 'pytorch_model_hub_mixin']" ] }, "execution_count": 29, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# Some other info\n", "card = ModelCard.load(\"daphne-cornelisse/policy_S10_000_02_27\")\n", "card.data.tags" ] }, { "cell_type": "code", "execution_count": 30, "metadata": {}, "outputs": [], "source": [ "# Model architecture\n", "#agent" ] }, { "cell_type": "code", "execution_count": 31, "metadata": {}, "outputs": [], "source": [ "# Weights \n", "#agent.state_dict()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Make environment" ] }, { "cell_type": "code", "execution_count": 32, "metadata": {}, "outputs": [], "source": [ "# Create data loader\n", "train_loader = SceneDataLoader(\n", " root='../../data/processed/examples',\n", " batch_size=num_envs,\n", " dataset_size=100,\n", " sample_with_replacement=False,\n", ")\n", "\n", "# Set params\n", "env_config = dataclasses.replace(\n", " EnvConfig(),\n", " ego_state=config.ego_state,\n", " road_map_obs=config.road_map_obs,\n", " partner_obs=config.partner_obs,\n", " reward_type=config.reward_type,\n", " norm_obs=config.norm_obs,\n", " dynamics_model=config.dynamics_model,\n", " collision_behavior=config.collision_behavior,\n", " dist_to_goal_threshold=config.dist_to_goal_threshold,\n", " polyline_reduction_threshold=config.polyline_reduction_threshold,\n", " remove_non_vehicles=config.remove_non_vehicles,\n", " lidar_obs=config.lidar_obs,\n", " disable_classic_obs=config.lidar_obs,\n", " obs_radius=config.obs_radius,\n", " steer_actions = torch.round(\n", " torch.linspace(-torch.pi, torch.pi, config.action_space_steer_disc), decimals=3 \n", " ),\n", " accel_actions = torch.round(\n", " torch.linspace(-4.0, 4.0, config.action_space_accel_disc), decimals=3\n", " ),\n", ")\n", "\n", "# Make env\n", "env = GPUDriveTorchEnv(\n", " config=env_config,\n", " data_loader=train_loader,\n", " max_cont_agents=config.max_controlled_agents,\n", " device=device,\n", ")" ] }, { "cell_type": "code", "execution_count": 33, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "['../../data/processed/examples/tfrecord-00000-of-01000_325.json',\n", " '../../data/processed/examples/tfrecord-00000-of-01000_4.json']" ] }, "execution_count": 33, "metadata": {}, "output_type": "execute_result" } ], "source": [ "env.data_batch" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Use the agent" ] }, { "cell_type": "code", "execution_count": 34, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "torch.Size([2, 64, 2984])" ] }, "execution_count": 34, "metadata": {}, "output_type": "execute_result" } ], "source": [ "next_obs = env.reset()\n", "\n", "control_mask = env.cont_agent_mask\n", "\n", "next_obs.shape" ] }, { "cell_type": "code", "execution_count": 35, "metadata": {}, "outputs": [], "source": [ "action, logprob, entropy, value = sim_agent(\n", " next_obs[control_mask], deterministic=False\n", ")" ] }, { "cell_type": "code", "execution_count": 36, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "(torch.Size([13]), torch.Size([13]), torch.Size([13]), torch.Size([13, 1]))" ] }, "execution_count": 36, "metadata": {}, "output_type": "execute_result" } ], "source": [ "action.shape, logprob.shape, entropy.shape, value.shape" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Rollout" ] }, { "cell_type": "code", "execution_count": 37, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "torch.Size([2, 64, 2984])\n", "Step: 54" ] } ], "source": [ "next_obs = env.reset()\n", "\n", "control_mask = env.cont_agent_mask\n", "\n", "print(next_obs.shape)\n", "\n", "frames = {f\"env_{i}\": [] for i in range(num_envs)}\n", "\n", "for time_step in range(env.episode_len):\n", " print(f\"\\rStep: {time_step}\", end=\"\", flush=True)\n", "\n", " # Predict actions\n", " action, _, _, _ = sim_agent(\n", " next_obs[control_mask], deterministic=False\n", " )\n", " action_template = torch.zeros(\n", " (num_envs, max_agents), dtype=torch.int64, device=device\n", " )\n", " action_template[control_mask] = action.to(device)\n", "\n", " # Step\n", " env.step_dynamics(action_template)\n", "\n", " # Render \n", " sim_states = env.vis.plot_simulator_state(\n", " env_indices=list(range(num_envs)),\n", " time_steps=[time_step]*num_envs,\n", " zoom_radius=70,\n", " )\n", " \n", " for i in range(num_envs):\n", " frames[f\"env_{i}\"].append(img_from_fig(sim_states[i])) \n", "\n", " next_obs = env.get_obs()\n", " reward = env.get_rewards()\n", " done = env.get_dones()\n", " info = env.get_infos()\n", " \n", " if done.all():\n", " break\n", "\n", "env.close()" ] }, { "cell_type": "code", "execution_count": 39, "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "
\n", "
env_0
\n", "
\n", "
env_1
" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "mediapy.show_videos(frames, fps=15, width=500, height=500, columns=2, codec='gif')" ] } ], "metadata": { "kernelspec": { "display_name": "gpudrive", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.11.9" } }, "nbformat": 4, "nbformat_minor": 2 }