Introduction
The mission of this thesis is to develop decentralized control algorithms that enable intelligent robots to interact with other robots or humans without explicit communication.
Thesis Title: Design and Analysis of Interactive Multi-Robot Systems without Explicit Communication.
This thesis tackles both cooperative and competitive interactions among robots. Chapter 2, Chapter 3, and Chapter 4 deal with cooperative tasks, while Chapter 5 is a fully competitive scenario.
Ch 2. Multi-Robot Manipulation via Force Consensus
We propose a decentralized force controller that enables a group of robots to transport a heavy object. Robots coordinate their actions by sensing the motion of the object, instead of explicit communication. Only one leader robot knows the desired trajectory and it can steer the rest of the group by varying its own force input. Extensive theoretical analysis is conducted to mathematically verify the convergence guarantee of the proposed controller under various scenarios and conditions.
The video below presents two simulations: (a) 12 robots transporting a rectangular plank; (b) 1000 robots transporting a grand piano of realistic weight and size.
Ch 3. Kinematic Wrench Coordination
We experimentally verify the multi-robot manipulation task using two types of custum-designed robot platforms. Our new kinematic wrench coordination approach shares the same underlying principles as in the previous chapter, but has some modification which is more amenable for physical implementation in real time and based on the particular robot hardware we have.
In the first video below, four differential-drive robots collaborate to transport a box. The robots are equipped with custom 2D force sensor and laser-based velocity sensor (for sensing the object's motion). A leader robot (or a human) in the front can guide the object's trajectory.
In the second video below, we use four omnidirectional robots to control not only translation, but also rotation of the payload.
Ch 4. Cooperative Aerial Manipulation
This chapter extends the scope of the two previous chapters to the 3D space. We attach many quadrotor aerial robots around the object to cooperatively lift the payload. We propose a novel distributed wrench controller and a trajectory optimization pipeline that accounts for the actuator limits, to guarantee the stability and feasibility of the entire system.
The video below shows a simulation of eight quadrotors delivering a heavy payload along a 3D trajectory in a cluttered environment.
We test our distributed wrench controller experimentally with two custom-built quadrotor robots, which collectively lift a payload without peer communication, shown in the video below.
Ch 5. Game Theoretic Multi-Robot Motion Planning
This chapter focuses on the competitive interactions among robots without communication. We present a real-time game theoretic planning algorithm for a robotic vehicle (e.g. a drone or a car) to race competitively against multiple opponents on a racecourse. Our algorithm plans receding horizon trajectories to maximally advance the robot along the racecourse, while taking into account the opponents’ intentions and responses. Our algorithm uses an iterative best response scheme to find approximate Nash equilibria in the space of the multiple robots’ trajectories. Rich behaviors such as blocking, overtaking, nudging or threatening emerge as a result of our algorithm, as can be seen in the experimental video below.
In the video, two slower robots starting in the front try to block the two faster robots starting in the back. The trajectories are planned independently by the robots in real time and are shown in the bottom right corner.
We also test our algorithm in 3D racecourse with onboard visual perception that is used to measure the relative position of the opponent robot in order to perform planning, as illustrated in the video below.