PA-SaR: Perception-Aware Learning, Control, and Coordination for Multi-Robot Search and Rescue
Organizations: Department of Intelligent Systems, Faculty of Electrical Engineering, Mathematics, and Computer Science, TU Delft, Van Mourik Broekmanweg 6, Delft, 2628 XE, Zuid-Holland, The Netherlands. · Delft Center for Systems and Control, Faculty of Mechanical Engineering, TU Delft, TU Delft, Mekelweg 2, Delft, 2628 CE, Zuid-Holland, The Netherlands.
Abstract
Autonomous SaR requires robots to explore unknown or partially known environments, avoid hazards, coordinate team activities, and guide discovered victims to safety. Improving individual capabilities of robots does not necessarily improve mission outcomes: safer motion can restrict the exploration behavior, faster discovery can increase the rescue demand, and effective guidance depends on victim responses. We introduce PA-SaR, a mathematical and architectural framework for coordinating heterogeneous local control, learning-based exploration, supervisory task allocation, and adaptive approach-lead victim guidance around mission-level objectives. PA- SaR distinguishes environmental ground-truth, robot-specific perceptions, and simulated victim behavior. This provides a common basis for integrating complementary mechanisms and analyzing their interactions across mission stages. We demonstrate the framework through single- and multi-robot simulation studies incorporating model-predictive and learning-based exploration, perception-based safety shielding, event-driven task allocation, and adaptive victim guidance. The results show trade-offs between collision avoidance and search completion, indicate that supervisory allocation improves search completion beyond information sharing alone, and demonstrate that increased exploration does not necessarily yield more rescues. Adaptive guidance increases rescues relative to continuous leading, without a detected improvement in mean or final victim trust or a reduction in mortality. These findings support evaluating local control, team coordination, and evacuation as related but distinct aspects of autonomous SaR within the considered simulation environments.
Figures & tables
| Variable | Description |
| Number of grid rows | |
| Number of grid columns | |
| Set of grid cells | |
| Discrete step | |
| Set of robots | |
| Set of active victims |
| Variable | Description |
| Robot position | |
| Robot perception state | |
| Set of admissible perception states | |
| Perceived traversability layer | |
| Perceived vulnerability layer | |
| Perceived victim layer |
| \small P_{r,ij}^{\mathrm{r,confidence}}(k+1)=\begin{cases}P_{r,ij}^{\mathrm{r,confidence}}(k)+\eta_{r,ij}(k+1)\left(1-P_{r,ij}^{\mathrm{r,confidence}}(k)\right),&\text{cell (i,j)k+1}\\ \max\left\{0,P_{r,ij}^{\mathrm{r,confidence}}(k)-\sigma_{r}^{\mathrm{confidence}}\right\},&\text{otherwise}.\end{cases} | (3) |
| Variable | Description |
| Victim state | |
| Victim position | |
| Victim age group | |
| Interaction strategy | |
| Victim trust | |
| Trust sensitivity |
| Interaction strategy | Autonomous preference when no danger is perceived | Conflict-resolution strategy |
| Altruistic | Approaches nearby victims | Yields to egoistic and selfish victims |
| Egoistic | Approaches nearby victims | Yields only to selfish victims |
| Selfish | Moves toward the exit | Takes priority over victims using other interaction strategies |
| Variable | Description |
| Predicted step | |
| Predicted robot position | |
| Candidate control sequence | |
| Predicted control input | |
| Set of length- input sequences | |
| Sequence objective value |
| Component | Definition |
| Discovery | |
| Confidence | |
| Vulnerability |
| Variable | Description |
| Transition reward | |
| Number of newly detected victims | |
| Terrain-collision increment | |
| Victim-collision increment | |
| Explored traversable-cell fraction | |
| Successful-completion indicator |
| Variable | Description |
| Candidate task type | |
| Robot task type | |
| Rescue execution status | |
| Frontier-component size | |
| Number of frontier targets | |
| Available robots at iteration |
| (13) |
| Variable | Description |
| Task-aware stage objective | |
| Robot-proximity score | |
| Path-progress score | |
| Perceived positions of other robots | |
| Nearest path-cell index | |
| Distance to nearest path cell |
| (14) |
| Component | Definition |
| Robot proximity | |
| Path progress |
| Controller | Completion | Collision-free |
| MPC | 97.2 | 50.2 |
| TMPC | 35.0 | 34.4 |
| SDD-TMPC | 81.6 | 81.2 |
| Completion | Collision-free | Runs with a collision | |
| SDD-TMPC | |||
| 0.0 | 82.4 | 82.4 | 0.0 |
| 0.1 | 84.8 | 84.8 | 0.0 |
| 0.2 | 78.8 | 78.8 | 0.0 |
| 0.4 | 76.8 | 75.2 | 3.6 |
| TMPC | |||
| Policy | Coverage | Recall | Success |
| Four-frame | |||
| Recurrent |
| Controller | Success | Coverage | Recall | Collisions | Harm | Time (ms) |
| MPC | 79 | 0.2 | 6.5 | 274 | ||
| TMPC | 57 | 0 | 11.9 | 129 | ||
| Pure RL | 69 | 117.7 | 175.8 | 12 | ||
| Shield | 64 | 0 | 12.0 | 40 | ||
| Shield+C | 64 | 0 | 12.1 | 53 |
| Architecture | Detection completion (%) | Duration (steps) |
| Supervised team | 91.7 | |
| Unsupervised team | 50.5 | |
| Shared team | 75.9 |
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
| Parameter/setting | Meaning and scope | Reference | Value/configuration |
|---|---|---|---|
| A. Environment generation | |||
| Room count | Number of generated rooms. | – | : ; : ; : ; : ; : at sizes ; : . |
| Room side bounds | Configured minimum/maximum width and length, in cells. | – | : ; : ; : ; : ; : . |
| Corridor-width bounds | Configured minimum/maximum corridor thickness. | – | : ; : ; : ; : . |
| Entrance dimensions | Starting-room width and length, in cells. | – | All scenarios: . |
| Room vulnerability probability | Probability that a room receives non-zero vulnerability. | – | : ; : ; : ; : ; : . |
Search-and-Rescue Robotics
Search-and-rescue robotics develops autonomous and human-guided robots to find, reach, assess, and assist people in hazardous environments.
See all 40 papers