Multi-robot cooperative search method in uncertain field environment

Abstract

The application discloses a kind of uncertainty field environment under multi-space-ground robot cooperative search method, it is related to multi-agent cooperative search technical field.The method includes: grid field task area, define space-ground robot state;Based on large language model, initialize the search value of task area cell;Design the total cost of cooperative search path planning including path cost and mode switching cost;Introduce communication distance constraint, adopt path decision method based on predictive control, select the minimum path point motion of cost, and dynamically update search value to carry out rolling optimization.The present application reduces invalid search area using large language model, improves search efficiency;Mode switching cost and different mode energy consumption are included in planning, avoid frequent switching and excessive dependence on high energy consumption mode, improve endurance.The present application is suitable for heterogeneous or homogeneous multi-agent system, significantly improves the practicability and efficiency of multi-space-ground robot cooperative search.

Description

Technical Field

[0002] This invention relates to the field of robot path planning, and in particular to a multi-air-ground robot cooperative search method in uncertain field environments. Background Technology

[0004] Air-to-ground robots can switch modes according to environmental conditions, enabling them to fly or move on the ground. This flexibility and vast operational space have garnered significant attention from academia and industry, and they are now applied in resource exploration, search and rescue, and environmental monitoring. However, a single air-to-ground robot often operates slowly and linearly when exploring unknown areas, especially in complex real-world scenarios such as the wilderness. Collaborative search using multiple air-to-ground robots can significantly improve search efficiency, meeting the high-efficiency requirements of tasks such as search and rescue in mountainous environments, fire tracing, and environmental protection.

[0005] A multi-aerial-ground robot system is a special type of multi-agent system. When all the aerial-ground robots maintain ground driving or flight mode, the multi-aerial-ground robot system can be regarded as a homogeneous multi-agent system. However, when the aerial-ground robot modes in the system are not uniform, that is, some robots fly and some robots drive on the ground, the multi-aerial-ground robot system evolves into a heterogeneous multi-agent system.

[0006] Currently, there are likely no research or case studies on multi-aerial-ground robot cooperative search in the scientific research and industrial fields. Traditional multi-agent cooperative search methods are mostly geared towards homogeneous and heterogeneous multi-agent systems, and research on multi-agent cooperative search for such variations is lacking. Extending traditional methods to multi-aerial-ground robot cooperative search can bring broader exploration possibilities to the research and application of air-ground robots, especially multi-aerial-ground robots. Summary of the Invention

[0008] To address the problems existing in the aforementioned background technology, this invention proposes a multi-aerial-ground robot cooperative search method for uncertain field environments. This method extends the traditional cooperative search methods applied to heterogeneous or homogeneous multi-aerial systems to dynamic multi-aerial systems, enabling multi-aerial-ground robots to improve search efficiency in field areas through mutual cooperation. Furthermore, it incorporates mode switching to enhance the practicality of cooperative planning and search.

[0009] This invention proposes a multi-air-ground robot cooperative search method for uncertain field environments, the steps of which include:

[0010] S1: Rasterized mission area E in the wild.

[0011] S2: Treating the air-to-ground robot as a point mass, defining... Time-Space-Ground Robot The state is ,in, , , and They represent Time-Space-Ground Robot The horizontal and vertical coordinate positions, pattern, and direction of motion. , This represents the number of air-to-ground robots in a multi-air-to-ground robot system.

[0012] S3: Initialize the task region based on a large language model Each cell Search value .

[0013] S4: Considering mode switching and the operating costs of different modes, design... Total cost of multi-space-ground robot collaborative search path planning in the wild environment for:

[0014]

[0015] In the formula, and These are path cost and mode switching cost, respectively.

[0016] S5: Introduces air-to-ground robot path decision-making that takes into account communication distance and is based on predictive control.

[0017] S5.1: Introduce communication distance constraints to set the maximum effective communication distance between individual air-to-ground robots. .

[0018] S5.2: Calculate the potential paths of each air-to-ground robot in the system. The cost.

[0019] S5.3: Each air-to-ground robot selects the path with the minimum cost. The first point is used as the next path point. And move to that point.

[0020] S5.4: Update Search Value Updated search value for:

[0021]

[0022] in, To maximize the search value, that is =100.

[0023] for:

[0024]

[0025] in, The value ranges from 0 to 1, and is a proportionality coefficient obtained through experience, with m being a coefficient greater than 1.

[0026] S5.5: Starting from the new position, repeat steps 5.2 to 5.5 to optimize the predicted path by rolling to make the air-to-ground robot plan the optimal route until the convergence condition is met. Here, the convergence condition is set to 350 time steps.

[0027] The advantages of this invention are:

[0028] (1) The multi-space-ground robot collaborative search method in uncertain field environment of the present invention combines expert evaluation and uses a large language model to provide initial search value for each grid, reduce invalid search areas, and improve the efficiency of multi-space-ground robot collaborative search.

[0029] (2) The multi-air-ground robot cooperative search method in uncertain field environments of the present invention takes into account the mode switching cost, avoids frequent mode switching of air-ground robots, and improves practicality;

[0030] (3) The multi-air-ground robot collaborative search method in uncertain field environment of the present invention takes into account the energy consumption of different modes at the same time, avoids excessive dependence on high energy consumption mode, and improves the endurance of air-ground robots.

[0031] (4) The present invention provides a collaborative search method for multi-space-ground robots in uncertain field environments. It designs a collaborative search method for multi-space-ground robots, provides an example for collaborative search of multi-space-ground robots, and is a cutting-edge and practical collaborative search method for multi-space-ground robots. Attached Figure Description

[0033] Figure 1 This is a flowchart of the multi-air-ground robot cooperative search method for uncertain field environments according to the present invention.

[0034] Figure 2 This is a task map with initial search value for the multi-space-ground robot cooperative search method in uncertain field environments according to the present invention.

[0035] Figure 3 This is a flowchart of the air-ground robot path decision-making method based on predictive control in the multi-air-ground robot cooperative search method under uncertain field environments of the present invention.

[0036] Figure 4 The graph shows the time-varying curve of the air-ground robot coverage rate of the multi-air-ground robot cooperative search method in uncertain field environments according to the present invention, and its comparison with the greedy algorithm. Detailed Implementation

[0038] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples.

[0039] This invention relates to a multi-air-ground robot cooperative search method for uncertain field environments, such as... Figure 1 As shown, the specific implementation steps include:

[0040] S1: The rasterized task area E in the field is a region composed of... A two-dimensional ordered pair composed of rectangular cells ;in, and These represent the horizontal and vertical indexes of the cell, respectively. Indicates the first Line number A rectangular cell. and These respectively indicate the task area in direction and The number of cells in the direction; in this embodiment, , There are a total of 1500 rectangular cells combined into two-dimensional ordered pairs.

[0041] S2: Treating the air-to-ground robot as a point mass, defining... Time-Space-Ground Robot The state is ,in, , , and They represent Time-Space-Ground Robot The horizontal and vertical coordinate positions, pattern, and direction of motion. , This refers to the number of air-to-ground robots in a multi-air-to-ground robot system. In this embodiment, the number of air-to-ground robots is 4. .

[0042] The above pattern ;in, Indicates flight mode. Indicates ground movement mode;

[0043] The direction of movement is:

[0044]

[0045] in, , , , , and These represent movement to the front, left front, right front, back, left, and right respectively.

[0046] S3: Initialize the task region based on a large language model Each cell Search value ;

[0047] Initial search value It is an integer between 0 and 100, represented as:

[0048]

[0049] in, Indicates expert opinion on the cell Natural language evaluation of search importance and Large language model and The parameters, and Representing large language models and The output of, where Used to transfer expert opinions to cells The natural language evaluation is converted into a numerical evaluation value according to preset rules, and the reasoning process is given. Used from Extract only the numerical evaluation values ​​from the output. and Large language model and Preset suggestion words. The initial search value of each cell is as follows: Figure 2 As shown.

[0050] Code snippet:

[0051] def init_values(grid: Grid, cfg: Dict) -> None:

[0052] vcfg = cfg.get("value_init", {})

[0053] notes_path = resolve_path(cfg, vcfg.get("expert_notes_path", ""))

[0054] notes = load_notes(notes_path)

[0055] llm_enabled = bool(vcfg.get("llm_enabled", False))

[0056] llm_client = None

[0057] if llm_enabled:

[0058] prompt_path = resolve_path(cfg, vcfg.get("prompt_path", ""))

[0059] prompt = prompt_path.read_text(encoding="utf-8") if prompt_path.exists() else ""

[0060] extract_prompt = ""

[0061] extract_path_raw = vcfg.get("extract_prompt_path", "")

[0062] if extract_path_raw:

[0063] extract_path = resolve_path(cfg, extract_path_raw)

[0064] if extract_path.exists():

[0065] extract_prompt = extract_path.read_text(encoding="utf-8")

[0066] llm_cfg = LLMConfig(

[0067] api_key_path=resolve_path(cfg, vcfg.get("api_key_path","")),

[0068] base_url = vcfg.get("base_url", ""),

[0069] model = vcfg.get("model", ""),

[0070] prompt = prompt,

[0071] extract_prompt = extract_prompt, )

[0073] llm_client = LLMClient(llm_cfg)

[0074] seed = int(vcfg.get("fallback_seed", 42))

[0075] for y in range(grid.height):

[0076] for x in range(grid.width):

[0077] note = note_for_cell(notes, x, y)

[0078] if llm_client is not None:

[0079] try:

[0080] score = llm_client.score_text(note)

[0081] except Exception:

[0082] score = heuristic_score(note, seed)

[0083] else:

[0084] score = heuristic_score(note, seed)

[0085] grid.values[y][x] = max(0.0, min(grid.max_value, float(score)))

[0086] The variable grid.values[y][x] in the above code represents the initial search value The `note_for_cell(notes, x, y)` function represents the expert's natural language evaluation of the cell. ;llm_client represents LLM; prompt, extract_prompt represent prompt words. and llm_client.score_text(note) indicates ;grid.height, grid.width means , ;grid.max_value represents the maximum limit for the value (100).

[0087] S4: Considering mode switching and the operating costs of different modes, design... Total cost of multi-space-ground robot collaborative search path planning in the wild environment for:

[0088]

[0089] In the formula, and These are path cost and mode switching cost, respectively.

[0090] Among them, path cost for:

[0091]

[0092] in, Indicates the weighting coefficient. The attenuation coefficient is... It is a natural constant; For coverage, This represents the total number of cells that have been searched at least once by an air-to-ground robot; Energy consumption coefficient The basic energy consumption per unit displacement. Multiples of experience; and This can be obtained through experimental calibration. In this embodiment, the energy consumption in the ground motion mode is set to unit energy consumption, i.e. =1, serving as a benchmark for energy consumption calculations, an empirical multiple. If we choose 10, then: . For feasible rectangular cells to air-to-ground robots Current location distance, For air-to-ground robots From current location Starting from the beginning, the next reachable rectangular cell is determined by the direction of movement. and It is confirmed that the cost of the robot's vertical movement has been summarized into Therefore, vertical displacement is no longer considered in distance calculations. Indicates the area surrounding the mission. The top, bottom, left, and right boundaries are the area borders. Indicates feasible rectangular cells to Distance to the boundary This represents the amount of angle change required to reach a feasible rectangular cell. The number of selectable motion directions, determined by the motion direction. and "Confirmed" can be expressed as:

[0093] ;

[0094] The aforementioned mode switching cost for:

[0095]

[0096] in, The energy consumption for switching from flight mode to ground mode serves as a benchmark for energy consumption during switching and can be determined empirically or experimentally; k is an empirical coefficient that is related to the robot's configuration; here, we set k=5. =1.2, then we have:

[0097] .

[0098] S4 code snippet:

[0099] def step_cost(

[0100] grid: Grid,

[0101] robot: Robot

[0102] x: int,

[0103] y: int,

[0104] angle_change: int,

[0105] mode: str,

[0106] params: CostParams,

[0107] ) -> float:

[0108] if robot.knowledge is None:

[0109] value = 0.0

[0110] else:

[0111] value = robot.knowledge.values[y][x]

[0112] value_term = params.value_weight * (1.0 - (value / grid.max_value))

[0113] coverage = grid.coverage_ratio()

[0114] coverage_term = math.exp(-params.coverage_decay * coverage)

[0115] energy = params.air_energy if mode == "air" else params.ground_energy

[0116] energy_term = params.energy_weight * energy

[0117] dist = grid.distance_to_boundary(x, y)

[0118] boundary_term = params.boundary_weight * (1.0 / (1.0 + dist))

[0119] turn_term = params.turn_weight * (angle_change / 180.0)

[0120] return coverage_term * value_term + energy_term + boundary_term +turn_term

[0121] def mode_switch_cost(robot: Robot, new_mode: str, params: CostParams)-> float:

[0122] if robot.mode == new_mode:

[0123] return 0.0

[0124] return params.switch_cost_k * params.switch_energy

[0125] The corresponding variable / expression in the code above is: step_cost() + mode_switch_cost(), which represents the total cost. step_cost() represents the path cost. mode_switch_cost(robot, new_mode, params) represents the mode switching cost. ; coverage = grid.coverage_ratio() means ; coverage_term = exp(-params.coverage_decay * coverage) means ; energy = params.air_energy if mode=='air' else params.ground_energy means .

[0126] S5: Introducing air-to-ground robot path decision-making based on predictive control and considering communication distance, such as... Figure 3 As shown.

[0127] S5.1: Introduce communication distance constraints.

[0128] When the distance between individual air-to-ground robots exceeds the communication limit, the search value of the task area to be exchanged will be lost.

[0129] Information, setting the maximum effective communication distance between individual air-to-ground robots. It is related to the communication equipment carried by the robot, and is set in this embodiment. The length of a single unit cell.

[0130] S5.2: Calculate the potential paths of each air-to-ground robot in the system. The cost.

[0131] Potential Path ,in The feasible path for flight mode Feasible paths for ground movement. This represents the total number of feasible paths; due to Typically shorter, and designed with mode switching costs. Air-to-ground robots typically do not switch modes frequently, therefore future developments are not considered. If a combination of flight and ground motion occurs within a step length, then the potential path for each air-to-ground robot in the system is... The cost is:

[0132]

[0133] in, and These are the costs of future planning and the weight of current planning. The predicted time step; in this embodiment and We selected 0.4 and 0.6 respectively, and T was empirically chosen to be 3. Index of time step, Representing a path Hollow-ground robot exist The search cost at any given moment is:

[0134]

[0135] in, express Time's up Time-Space-Ground Robot The average predicted search cost.

[0136] S5.3: Each air-to-ground robot selects the path with the minimum cost. The first point is used as the next path point. And move to that point. The code snippet for S5.3 is as follows:

[0137] def best_next_step(

[0138] robot: Robot

[0139] grid: Grid,

[0140] horizon: int,

[0141] w_current: float,

[0142] w_future: float,

[0143] params: CostParams,

[0144] ) -> Tuple[int, int, int, str]:

[0145] best: PathOption | None = None

[0146] for mode in ("ground", "air"):

[0147] options = enumerate_paths(robot, mode, horizon, grid)

[0148] options = evaluate_paths(robot, grid, options, horizon, w_current, w_future, params)

[0149] for opt in options:

[0150] if best is None or opt.cost < best.cost:

[0151] best = opt

[0152] if best is None or not best.positions:

[0153] return robot.x, robot.y, robot.heading, robot.mode

[0154] nx, ny = best.positions[0]

[0155] nheading = best.headings[0]

[0156] return nx, ny, nheading, best.mode

[0157] The corresponding variables / expressions in the above code are: w_current, w_future represent w_h,g; opt.cost represents C_k,g(w_h,g); params.value_weight, params.coverage_decay represent α1, α2; and horizon represents T.

[0158] S5.4: Update Search Value Updated search value for:

[0159]

[0160] in, To maximize the search value, that is =100.

[0161] for:

[0162]

[0163] in, The value ranges from 0 to 1, and is a proportionality coefficient obtained through experience. The coefficient is greater than 1; therefore, the above formula can be expressed as follows: after the current area has been searched, the value of searching again decreases, and ground-based methods are better at exploring the searched area than air-based methods. The code snippet is as follows:

[0164] def apply_visit(self, x: int, y: int, mode: str, theta: float, m:float) -> None:

[0165] if not self.in_bounds(x, y): return

[0166] self.visits[y][x] += 1

[0167] k = self.visits[y][x]

[0168] p_initial = self.initial_values[y][x]

[0169] factor = theta if mode in ("air", "flying") else (m * theta ifmode in ("ground", "driving") else None)

[0170] if factor is None: raise ValueError(f"Unknown mode: {mode}")

[0171] If factor < 0.0: factor = 0.0

[0172] If factor > 1.0: factor = 1.0

[0173] p_update = p_initial * (factor ** k)

[0174] if p_update < 0.0: p_update = 0.0

[0175] if p_update > self.max_value: p_update = self.max_value

[0176] self.values[y][x] = p_update

[0177] S5.5: Starting from the new position, repeat steps 5.2 to 5.5 to optimize the predicted path by rolling to make the air-to-ground robot plan the optimal route until the convergence condition is met. Here, the convergence condition is set to 350 time steps.

[0178] Based on the multi-air-ground robot cooperative search method for uncertain field environments proposed in this invention, when multi-air-ground robots conduct cooperative searches... All identified regions with an initial value greater than 50 (which can be considered high-value regions) were searched at least once. For example... Figure 4 The figure shows the coverage rate of the method of the present invention over time and its comparison with that of the greedy search algorithm. As can be seen from the figure, the coverage rate of the method of the present invention far exceeds that of the greedy algorithm for most step lengths. The difference between the two algorithms is smallest at about 200 step lengths, but the method still maintains the lead.

Claims

1. A multi-air-ground robot cooperative search method for uncertain field environments, characterized by: steps include: S1: Rasterized mission area E in the field; S2: Treating the air-to-ground robot as a point mass, defining... Time-Space-Ground Robot The state is ,in, , , and They represent Time-Space-Ground Robot The horizontal and vertical coordinate positions, pattern, and direction of motion. , This refers to the number of air-to-ground robots in a multi-air-to-ground robot system. S3: Initialize the task region based on a large language model Each cell Search value ; S4: Considering mode switching and the operating costs of different modes, design... Total cost of multi-space-ground robot collaborative search path planning in the wild environment for: ； In the formula, and These are path cost and mode switching cost, respectively. S5: Introducing air-to-ground robot path decision-making based on predictive control and considering communication distance; S5.1: Introduce communication distance constraints to set the maximum effective communication distance between individual air-to-ground robots. ; S5.2: Calculate the potential paths of each air-to-ground robot in the system. The cost; S5.3: Each air-to-ground robot selects the path with the minimum cost. The first point is used as the next path point. and move to that point; S5.4: Update Search Value Updated search value for: ； in, To maximize the search value, that is =100; for: ； in, The value ranges from 0 to 1, representing a proportionality coefficient obtained through experience, and m is a coefficient greater than 1; S5.5: Starting from the new position, repeat steps 5.2 to 5.5 to optimize the predicted path by rolling to make the air-to-ground robot plan the optimal route until the convergence condition is met. Here, the convergence condition is set to 350 time steps.

2. The multi-air-ground robot cooperative search method in an uncertain field environment as described in claim 1, characterized in that: In step S1, the task area E is gridded into... A two-dimensional ordered pair composed of rectangular cells ;in, and These represent the horizontal and vertical indexes of the cell, respectively.

3. The multi-air-ground robot cooperative search method in an uncertain field environment as described in claim 1, characterized in that: In step S2, the pattern ;in, Indicates flight mode. Indicates ground movement mode; The direction of movement is: ； in, , , , , and These represent movement to the front, left front, right front, back, left, and right respectively.

4. The multi-air-ground robot cooperative search method in an uncertain field environment as described in claim 1, characterized in that: In step S3, the initial search value It is an integer between 0 and 100, represented as: ； in, Indicates expert opinion on the cell Natural language evaluation of search importance and Large language model and The parameters, and Representing large language models and The output, Used to transfer expert opinions to cells The natural language evaluation is converted into a numerical evaluation value according to preset rules, and the reasoning process is given. Used from Extract only the numerical evaluation values ​​from the output. and Large language model and Preset prompts.

5. The multi-air-ground robot cooperative search method in an uncertain field environment as described in claim 1, characterized in that: In step S4, path cost for: ； in, Indicates the weighting coefficient. The attenuation coefficient is... It is a natural constant; For coverage, This represents the total number of cells that have been searched at least once by an air-to-ground robot; Energy consumption coefficient The basic energy consumption per unit displacement. As an experience multiple, Indicates flight mode. Indicates ground movement mode; For feasible rectangular cells to air-to-ground robots Current location distance, For air-to-ground robots From current location Starting point, the next reachable rectangular cell; Indicates the area surrounding the mission. The four boundaries are: top, bottom, left, and right. Indicates feasible rectangular cells to Distance to the boundary This represents the amount of angle change required to reach a feasible rectangular cell. The number of selectable motion directions, determined by the motion direction. and "Confirmed" can be expressed as: 。 6. The multi-air-ground robot cooperative search method in an uncertain field environment as described in claim 1, characterized in that: In step S4, the mode switching cost for: ； in, The energy consumption for switching from flight mode to ground mode serves as a benchmark for energy consumption during switching and can be determined empirically or experimentally; k is an empirical coefficient that is related to the robot's configuration.

7. The multi-air-ground robot cooperative search method in an uncertain field environment as described in claim 1, characterized in that: In step S5.2, the potential path ,in The feasible path for flight mode Feasible paths for ground movement. Let be the total number of feasible paths; then the potential paths for each air-to-ground robot in the system are... The cost is: ； in, and These are the costs of future planning and the weight of current planning. For the predicted time step, Index of time step, Representing a path Hollow-ground robot exist The search cost at any given moment is: ； in, express Time's up Time-Space-Ground Robot The average predicted search cost.

Images