Commit Graph

16 Commits

Author SHA1 Message Date
gqt
524ca8c070 Avoid wall-hugging during unknown recharge routes 2026-04-26 20:33:51 +08:00
gqt
69b8a692db Improve PPO diagnostics and recharge behavior 2026-04-26 20:24:26 +08:00
gqt
5b6133db13 Optimize PPO coverage and recharge strategy 2026-04-26 19:25:05 +08:00
gqt
220de372e0 调整PPO奖励突出有效充电 2026-04-26 18:56:42 +08:00
gqt
e99a224d86 优化PPO自适应回充与泛化特征 2026-04-26 18:35:23 +08:00
gqt
00b26af3ed 增加行为监控指标 2026-04-26 17:42:30 +08:00
gqt
5c2df10150 修复低电量回充卡住 2026-04-26 17:37:17 +08:00
gqt
f44e2483fc 优化 PPO 清扫策略 2026-04-26 17:29:03 +08:00
gqt
f04feb0cd9 增加PPO回充安全动作约束 2026-04-26 17:06:54 +08:00
gqt
e0756b4846 调整PPO回充模式清扫与探索奖励 2026-04-26 16:33:44 +08:00
gqt
3c3332e126 优化PPO基于电量安全余量回充 2026-04-26 16:20:02 +08:00
gqt
3d0a8122bb 修复PPO评估推理返回None异常 2026-04-26 15:35:19 +08:00
gqt
ba6cf2a797 修正PPO充电奖励防止蹲桩 2026-04-26 15:08:43 +08:00
gqt
efbc612945 优化PPO充电与避障策略
扩展观测特征到157维,加入充电桩、NPC、电量安全余量、地图统计和本步清扫信息。

增加低电量回充动作过滤、NPC危险区过滤,并调整奖励和终局日志以突出充电、避障和真实清扫得分。
2026-04-26 14:14:18 +08:00
gqt
eb3efa4df7 Optimize PPO short-run training 2026-04-26 12:46:00 +08:00
gqt
ca6234c941 Initial robot vacuum code 2026-04-26 12:38:39 +08:00