Search

CN-121995920-A - Dynamic prediction-based mobile robot indoor autonomous exploration hierarchical planning method

CN121995920ACN 121995920 ACN121995920 ACN 121995920ACN-121995920-A

Abstract

The invention discloses a mobile robot indoor autonomous exploration hierarchical planning method based on dynamic prediction, which comprises the following steps of S1 dividing an exploration environment where a robot is located into a plurality of local subspaces, marking the state of each subspace as unexplored, explored or semi-explored according to sensor data and semantic detection results, S2 establishing a global topological graph connected with each subspace, planning a global access path by using a travel business problem model, S3 integrating a pedestrian track prediction model in a selected current subspace, evaluating and selecting a local viewpoint capable of maximizing information gain and minimizing dynamic collision risk based on the prediction results, and the method has the advantages that the map is eliminated, the completeness of exploration is improved, the pedestrian track prediction is incorporated into the viewpoint, and the deadlock of a crowd is avoided by introducing a semi-explored state and a self-adaptive revising mechanism.

Inventors

  • WANG JUN
  • XU XINNING

Assignees

  • 北京化工大学

Dates

Publication Date
20260508
Application Date
20260204

Claims (9)

  1. 1. The mobile robot indoor autonomous exploration hierarchical planning method based on dynamic prediction is characterized by comprising the following steps of: S1, dividing an exploration environment where a robot is located into a plurality of local subspaces, and marking the state of each subspace as unexplored, explored or semi-explored according to sensor data and semantic detection results, wherein the semi-explored state refers to an area which is convenient to explore and is currently identified as being blocked by a movable object; S2, establishing a global topological graph connected with all subspaces, and planning a global access path by using a travel business problem model, wherein when the path cost reaching the subspace in a semi-exploration state is calculated, dynamic weight based on time attenuation and environmental triggering is introduced for correction; S3, integrating a pedestrian track prediction model in the selected current subspace, and evaluating and selecting a local viewpoint capable of maximizing information gain and minimizing dynamic collision risk based on a prediction result; s4, generating a smooth track according to the selected viewpoint sequence and controlling the robot to execute.
  2. 2. The mobile robot indoor autonomous exploration hierarchical planning method based on dynamic prediction of claim 1, wherein the semi-exploration state movable object comprises a door, a chair or a cart, and the semi-exploration state subspace has a first discovered time stamp.
  3. 3. The method according to claim 1, wherein in step S2, the distance cost from the current node i to the node j in the semi-exploration state is calculated Cost after correction Wherein the weight α is a function based on time decay and environmental triggering: , wherein, And when the pedestrian is detected to pass through the shielding object, the weight is rapidly reduced, and the global re-planning is triggered.
  4. 4. A mobile robot indoor autonomous exploration hierarchical planning method based on dynamic prediction as claimed in claim 3, wherein said state transition index The calculation formula of (2) is as follows: , Wherein, the The current position of the robot; The method is characterized in that the method comprises the steps of semi-exploring the mass center of subspace, wherein t is the time length from the first discovery to the present, gamma is a directivity indication factor, theta is the deviation angle between the direction of a robot and a target, xi is a pedestrian interaction zone bit, gamma is a space proximity indication factor, lambda is a large constant for triggering state switching, and beta and delta are adjustment coefficients for balancing the weights of all factors.
  5. 5. The method according to claim 4, wherein when the semantic perception system detects that a pedestrian enters the semi-exploration subspace or interacts with a movable object in the area, the pedestrian interaction flag bit is set to 1, so that the weight alpha quickly approaches to 1, and the global planner is triggered to preferentially incorporate the subspace into an access sequence.
  6. 6. The method is characterized in that a pedestrian track prediction model in the step S3 adopts a neural network based on an encoder-decoder structure, inputs of the neural network comprise a robot state, a local occupation grid map and the historical track of surrounding pedestrians, and outputs a speed vector sequence for a period of time in the future of the pedestrians, and the prediction model adopts a self-supervision continuous learning mechanism and utilizes the prediction and observation errors to finely adjust network parameters on line.
  7. 7. The method for hierarchical planning of autonomous indoor exploration of mobile robot based on dynamic prediction according to claim 1, wherein in step S3, a reward function including dynamic collision penalty is adopted To evaluate candidate viewpoint i, wherein Is the information gain; a collision risk calculated for a sequence based on the predicted pedestrian speed; And Is a weight coefficient.
  8. 8. The dynamic prediction-based mobile robot indoor autonomous exploration hierarchical planning method according to claim 7, wherein the collision risk is Based on a speed obstacle model, using a predicted future speed sequence of pedestrians For viewpoint i, it is assumed that the robot is at speed Forward, predicted collision time , , Wherein, the The method comprises the steps of calculating the position of a pedestrian at tau according to a prediction model, and calculating a punishment term if collision risk exists: , Wherein, the To estimate the arrival time; for a predicted collision time and if there is no risk of collision, 。
  9. 9. The method for planning the indoor autonomous exploration hierarchy of the mobile robot based on the dynamic prediction according to claim 1 is characterized in that in the step S5, a smooth track is generated by using a B-spline curve and is executed by combining a model prediction controller, and in the execution process, a reactive obstacle avoidance based on a gradient speed obstacle is adopted as safety redundancy.

Description

Dynamic prediction-based mobile robot indoor autonomous exploration hierarchical planning method Technical Field The invention relates to the cross field of mobile robot technology, artificial intelligence and autonomous navigation technology, in particular to a mobile robot indoor autonomous exploration hierarchical planning method based on dynamic prediction. Background Autonomous exploration (Autonomous Exploration) is one of the core capabilities of mobile robots to achieve intelligent operation, and aims to actively perceive an unknown environment through an onboard sensor (such as a laser radar LiDAR and a depth camera RGB-D) under the condition that no prior map information exists, and plan a motion path to maximally acquire environment information until a complete environment map is constructed. The technology has irreplaceable value in the fields of disaster relief (such as internal search and rescue of collapsed buildings), alien detection (such as mars navigation), underground mine exploration, indoor service (such as market cleaning and hospital material distribution) and the like. The current autonomous exploration methods are mainly classified into a boundary-based method, a Sampling-based method, and a hierarchical planning method combining the two methods (such as TARE PLANNER, FAEL). However, the prior art faces two major core challenges in practical applications, resulting in poor performance in dynamic indoor environments: 1. Most of the existing exploration algorithms assume that the environment is static. When a robot encounters a dynamic obstacle such as a pedestrian, passive reaction is usually carried out only by means of a local obstacle avoidance algorithm (such as DWA) on the bottom layer. The planning layer lacks the predictive capability of future motion trail of pedestrians, so that the selected view point (ViewPoint) of the robot is often blocked by the future path of the pedestrians, and the robot is forced to frequently stop suddenly, rotate in place or sink into crowd surrounding, so that the exploration efficiency is greatly reduced. There are a large number of "movable objects" (Movable Objects) in the indoor environment, such as doors, chairs, carts, etc. Existing algorithms typically label scanned closed doors directly as permanent obstacles (walls) and permanently cull the area behind the door from the exploration map (Graph). Even if the door is opened later, the robot cannot revisit the area automatically because the node has been abandoned in the internal topology. This results in map construction imperfections that are unacceptable in search and rescue or security tasks that require extremely high coverage. Therefore, an autonomous exploration and planning method capable of actively predicting pedestrian dynamics and intelligently processing the state change of a movable object is needed. Disclosure of Invention The invention aims to provide a mobile robot indoor autonomous exploration hierarchical planning method based on dynamic prediction, which aims to solve the problems of low exploration efficiency and insufficient coverage rate in dynamic and semi-static environments in the prior art. In order to achieve the above purpose, the mobile robot indoor autonomous exploration hierarchical planning method based on dynamic prediction comprises the following steps: S1, dividing an exploration environment where a robot is located into a plurality of local subspaces, and marking the state of each subspace as unexplored, explored or semi-explored according to sensor data and semantic detection results, wherein the semi-explored state refers to an area which is convenient to explore and is currently identified as being blocked by a movable object; S2, establishing a global topological graph connected with all subspaces, and planning a global access path by using a travel business problem model, wherein when the path cost reaching the subspace in a semi-exploration state is calculated, dynamic weight based on time attenuation and environmental triggering is introduced for correction; S3, integrating a pedestrian track prediction model in the selected current subspace, and evaluating and selecting a local viewpoint capable of maximizing information gain and minimizing dynamic collision risk based on a prediction result; s4, generating a smooth track according to the selected viewpoint sequence and controlling the robot to execute. Preferably, the semi-exploring movable object comprises a door, chair or cart, and the semi-exploring subspace has a first discovered time stamp. Preferably, in the step S2, a distance cost from the current node i to the node j in the semi-explored stateCost after correctionWherein the weight α is a function based on time decay and environmental triggering: , wherein, And when the pedestrian is detected to pass through the shielding object, the weight is rapidly reduced, and the global re-planning is triggered. Preferably, the state transition ind