Fix VBot navigation to use unified target positions for arena scoring

[Copilot]

VBot arena navigation was generating per-robot random targets (robot_init_pos + random_offset), making unified scoring impossible. All robots now navigate to the same fixed target.

Changes

Configuration (cfg.py)

• Added target_point_a = [0.0, 1.5] (inner circle trigger)
• Added target_point_b = [0.0, 0.0] (center, for future multi-stage)

Reset logic (vbot_section001_np.py)

# Before: each robot gets different target
target_positions = robot_init_pos[:, :2] + sampled[:, :2]

# After: all robots get unified target
target_point_a = np.array(cfg.target_point_a, dtype=np.float32)
target_positions = np.tile(target_point_a + arena_center, (num_envs, 1))

Target heading remains randomized for training diversity. Backward compatible with hasattr() checks.

Original prompt

问题描述

VBot 竞技场导航任务中，导航目标位置设定不正确，导致机器狗无法正确导航到圆心。

现状 ❌

• 目标位置计算: target_positions = robot_init_pos[:, :2] + sampled[:, :2]
• 问题: 目标位置 = 初始位置 + 随机偏移 [-5.0~5.0m]
• 结果: 每只狗的目标位置都不同，且不在圆心，无法统一计分
• 表现: 机器狗导航方向随机，无法完成"外圈→内圈→圆心"的三阶段任务

要求 ✅

根据题目和竞技场规范：

• 所有机器狗应该有统一的、固定的导航目标
• 阶段1: 从外圈 → 导航到内圈触发点 (0.0, 1.5)
• 阶段2: 从内圈 → 导航到圆心 (0.0, 0.0)
• 最终目标必须是圆心 (0.0, 0.0)

修复范围

1. cfg.py: 添加竞技场导航参数配置
2. vbot_section001_np.py - reset(): 修改目标位置生成逻辑，使用固定的触发点而非随机偏移
3. vbot_section001_np.py - step(): 实现三阶段导航逻辑（如果需要）

技术细节

需要添加到 cfg.py

# VBotSection001EnvCfg 配置类中添加：
arena_center: list = field(default_factory=lambda: [0.0, 0.0])
arena_outer_radius: float = 3.0
arena_inner_radius: float = 1.5
boundary_radius: float = 3.5
target_point_a: list = field(default_factory=lambda: [0.0, 1.5])   # 内圈触发点（+1分）
target_point_b: list = field(default_factory=lambda: [0.0, 0.0])   # 圆心（+1分）

需要修改的代码（reset 方法）

当前代码位置: motrix_envs/src/motrix_envs/navigation/vbot/vbot_section001_np.py 第 795-815 行

当前逻辑:

cmd_range = cfg.commands.pose_command_range
if len(cmd_range) != 6:
    raise ValueError("commands.pose_command_range must have 6 values...")
    
dx_min, dy_min, yaw_min, dx_max, dy_max, yaw_max = cmd_range
sampled = np.random.uniform(
    low=np.array([dx_min, dy_min, yaw_min], dtype=np.float32),
    high=np.array([dx_max, dy_max, yaw_max], dtype=np.float32),
    size=(num_envs, 3),
)
target_positions = robot_init_pos[:, :2] + sampled[:, :2]  # ❌ 错误
target_headings = sampled[:, 2:3]
pose_commands = np.concatenate([target_positions, target_headings], axis=1)

正确逻辑:

# 竞技场模式：固定目标为内圈触发点 A
target_point_a = np.array(cfg.target_point_a if hasattr(cfg, 'target_point_a') else [0.0, 1.5], dtype=np.float32)
arena_center = np.array(cfg.arena_center if hasattr(cfg, 'arena_center') else [0.0, 0.0], dtype=np.float32)

# 所有机器狗的目标统一为内圈触发点 A
target_positions = np.tile(target_point_a + arena_center, (num_envs, 1))

# 朝向仍然随机（或也可以固定为 0）
if hasattr(cfg.commands, 'pose_command_range') and len(cfg.commands.pose_command_range) == 6:
    yaw_min, yaw_max = cfg.commands.pose_command_range[2], cfg.commands.pose_command_range[5]
    target_headings = np.random.uniform(yaw_min, yaw_max, size=(num_envs, 1))
else:
    target_headings = np.zeros((num_envs, 1), dtype=np.float32)

pose_commands = np.concatenate([target_positions, target_headings], axis=1)

验证标准

• 初始化后所有机器狗的目标位置相同，都在 (0.0, 1.5) 或圆心 (0.0, 0.0)
• 目标位置在可视化中正确显示（绿色标记应该在同一位置）
• 机器狗能够导航向目标点
• 到达触发点时计分系统能正常工作

This pull request was created from Copilot chat.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.